Infrared Spectroscopy in Food Analysis: A Comprehensive Guide to Quality Control and Authenticity Testing

Abstract

This article provides a comprehensive overview of the application of infrared (IR) spectroscopy for ensuring food quality and authenticity. Tailored for researchers, scientists, and development professionals, it explores the foundational principles of IR spectroscopy, including Near-Infrared (NIR) and Fourier-Transform Infrared (FTIR) techniques. It details methodological approaches for analyzing diverse food matrices—from plant-based beverages and spices to nuts and dairy—highlighting the critical role of chemometrics for data analysis. The content further addresses practical challenges and optimization strategies for method development and compares IR spectroscopy's performance against traditional analytical techniques, offering insights into its validation, accuracy, and growing potential in industrial and research settings.

The Science of Light: Foundational Principles of Infrared Spectroscopy in Food Analysis

Infrared (IR) spectroscopy has emerged as a cornerstone analytical technique for ensuring food quality, safety, and authenticity. This vibrational spectroscopy method analyzes the interaction of electromagnetic radiation with matter to reveal detailed information about molecular composition and structure. Within the food sector, two specific regions of the infrared spectrum are predominantly utilized: the Near-Infrared (NIR) and Mid-Infrared (MIR) regions. The application of these techniques is particularly valuable for addressing critical challenges in food analysis, including the detection of adulteration, confirmation of geographical origin, quantification of key components, and identification of contaminants. Unlike traditional wet chemistry methods which often require extensive sample preparation, hazardous chemicals, and significant time, IR spectroscopy offers a rapid, non-destructive, and environmentally friendly alternative [1] [2]. The integration of chemometrics—the application of mathematical and statistical methods to chemical data—has further empowered researchers to extract meaningful information from complex spectral data, solidifying the role of IR spectroscopy as an indispensable tool in modern food science [2] [3].

Core Principles of Molecular Vibrations

The Physical Basis of Vibrational Spectroscopy

The fundamental principle underlying infrared spectroscopy is the absorption of specific frequencies of infrared light by chemical bonds within a molecule. This absorption occurs when the frequency of the incident IR radiation matches the natural vibrational frequency of a molecular bond, causing it to stretch, bend, or deform. The absorbed energy promotes the molecule to a higher vibrational energy state. The specific frequencies at which absorption occurs, measured in wavenumbers (cm⁻¹), provide a characteristic molecular fingerprint that is unique to the sample's chemical composition [4] [5].

Characteristic Vibrations in NIR and MIR

While both NIR and MIR spectroscopy probe molecular vibrations, they access different types of transitions, leading to distinct spectral information and applications.

Mid-Infrared (MIR) Spectroscopy operates in the range of 4000–400 cm⁻¹ (2.5–25 µm) and is concerned with the fundamental vibrations of molecular bonds. These include both stretching (symmetrical and asymmetrical) and bending (scissoring, rocking, wagging, twisting) motions. The MIR spectrum is divided into two key regions: the functional group region (4000–1300 cm⁻¹) and the fingerprint region (1300–600 cm⁻¹). The functional group region contains distinct absorption bands for key groups like O-H, N-H, and C-H, while the fingerprint region is characterized by complex, unique patterns resulting from coupled vibrations, allowing for precise material identification [4].

Near-Infrared (NIR) Spectroscopy covers the range of 12,800–4000 cm⁻¹ (780–2500 nm). This region contains signals from overtones and combinations of the fundamental vibrations observed in the MIR region. Specifically, NIR spectra arise from the first, second, and third overtones of X-H stretching modes (where X is C, N, or O) and combination bands (e.g., stretching + bending). These transitions are weaker by an order of magnitude compared to fundamental bands, resulting in NIR spectra with broad, overlapping peaks that are highly complex and difficult to interpret visually without multivariate statistical tools [2] [3].

Table 1: Comparison of NIR and MIR Spectroscopy

Feature	Near-Infrared (NIR)	Mid-Infrared (MIR)
Spectral Range	780–2500 nm (12,800–4,000 cm⁻¹)	2.5–25 µm (4,000–400 cm⁻¹)
Vibrational Transitions	Overtones & combinations	Fundamental vibrations
Absorption Intensity	Weaker (10-100x less than MIR)	Stronger
Sample Penetration	Higher (several mm)	Lower (micrometers)
Typical Sampling	Diffuse reflection, transmission	ATR, Transmission, DRIFT
Primary Bonds Probed	C-H, O-H, N-H	C=O, O-H, N-H, C-O, C-H

Spectral Interpretation and Data Analysis

Pre-processing of Spectral Data

Raw spectral data is often contaminated with noise and light scattering effects that can obscure chemical information. Therefore, data pre-processing is a critical first step in analysis.

• Multiplicative Scatter Correction (MSC): This is a model-based method that corrects for both additive and multiplicative effects commonly found in diffuse reflectance spectroscopy, effectively removing scattering artifacts [2].
• Standard Normal Variate (SNV): This technique centers and scales spectral data on a sample-by-sample basis, correcting for baseline shifts and path length variations [2].
• Savitzky-Golay Smoothing and Derivatives: This algorithm is widely used for smoothing spectra to reduce high-frequency noise. Its derivative function (first or second derivative) is applied to resolve overlapping absorption bands and correct baseline drift, though it may increase noise [2].

Chemometric Methods for Qualitative and Quantitative Analysis

Due to the complexity of IR spectra, especially in NIR, chemometric tools are essential for extracting meaningful information.

• Principal Component Analysis (PCA): An unsupervised technique used for exploratory data analysis. PCA reduces the dimensionality of the spectral data, helping to identify patterns, groupings, or outliers without prior knowledge of sample classes [2] [6].
• Partial Least Squares Regression (PLSR): A supervised method that is the workhorse for quantitative analysis. PLSR develops a model that finds the relationship between spectral data (X-matrix) and reference laboratory values (Y-matrix) for a constituent of interest (e.g., protein, moisture) [7] [2].
• Classification Methods (PLS-DA, SVM, Ensemble Learning): For qualitative analysis, such as authenticating origin or detecting adulteration, classification models are used. Partial Least Squares Discriminant Analysis (PLS-DA) is a linear classification method. More advanced non-linear methods like Support Vector Machine (SVM) and ensemble methods like Gradient Boosting Machine (GBM) can achieve higher accuracy for complex discrimination tasks [7] [6]. Ensemble learning, which combines multiple models, has shown great promise in improving predictive performance and robustness [7].

Experimental Protocols

Protocol 1: FT-NIR with Ensemble Learning for Detecting Contaminants in Oils

This protocol outlines the procedure for detecting mineral oil adulteration in corn oil using Fourier Transform Near-Infrared (FT-NIR) spectroscopy coupled with interpretable ensemble learning models [7].

1. Reagent and Sample Preparation:

• Acquire pure food-grade corn oil and potential contaminants (e.g., diesel, kerosene, lubricating oil).
• Prepare adulterated samples by spiking pure corn oil with individual mineral oils at multiple concentration levels (e.g., 0.5%, 1%, 2%, 5% v/v).
• Ensure homogeneous mixing for each sample.

2. Spectral Data Acquisition:

• Use an FT-NIR spectrometer equipped with a transmission cell or an ATR accessory.
• Collect spectra in the range of 800–2500 nm (12,500–4,000 cm⁻¹).
• For each sample, acquire multiple scans (e.g., 32–64) and average them to improve the signal-to-noise ratio.
• Maintain a constant temperature during measurement.

3. Data Pre-processing and Partitioning:

• Apply pre-processing techniques such as SNV, first derivative (Savitzky-Golay), or MSC to the raw spectra.
• Implement a clear data partitioning strategy. For each contamination level, randomly assign a portion of samples (e.g., 70%) to the training set for model development and the remainder (e.g., 30%) to the test set for validation.

4. Model Development and Validation:

• Develop a PLS-DA model as a baseline to screen contaminated vs. uncontaminated samples.
• Train ensemble learning models, such as LightGBM or Gradient Boosting Machine (GBM), to identify the specific type of contaminant.
• Optimize model hyperparameters using the training set via cross-validation.
• Validate the final model's performance on the untouched test set by evaluating metrics such as classification accuracy, sensitivity, and specificity.

5. Model Interpretation:

• Employ interpretation frameworks like SHAP (Shapley Additive Explanations) to identify the key spectral variables (wavelengths) that most significantly contribute to the model's predictions. This enhances trust in the model by providing insights into the chemical basis for the discrimination [7].

Protocol 2: Solvent-Based MIR for Authenticating Saffron

This protocol describes the use of solvent-based MIR spectroscopy for the geographical discrimination of saffron and detection of plant-based adulterants [6].

1. Metabolite Extraction:

• Follow a modified solvent extraction protocol based on the ISO 3632 standard for saffron.
• Use a Design of Experiments (DOE) approach to screen and optimize extraction parameters such as solvent type, extraction time, and temperature.

2. Spectral Acquisition with MIRA Analyzer:

• Utilize a dedicated MIR analyzer (e.g., MIRA) equipped with an ATR crystal.
• Place a droplet of the saffron extract onto the ATR crystal.
• Collect the MIR spectrum in the fingerprint region. Acquire multiple scans per sample to ensure representativeness.

3. Data Analysis for Geographical Discrimination:

• Perform exploratory data analysis using PCA to observe natural clustering of samples by origin.
• Develop a PLS-DA model using the MIR data to classify samples based on their geographical origin. Test different pre-processing methods (e.g., first derivative) to improve model performance.
• Alternatively, employ a Random Subspace Discriminant Ensemble (RSDE) model, which may achieve higher accuracy without intensive pre-processing.

4. Adulteration Detection and Identification:

• For adulterant detection, use a one-class model like Data-Driven Soft Independent Modeling of Class Analogy (DD-SIMCA) to differentiate pure saffron from adulterated samples.
• To identify the type and level of specific adulterants (e.g., safflower, marigold), build a multi-class classification model using PLS-DA or RSDE. The RSDE model has been shown to outperform PLS-DA for this task [6].

Diagram 1: MIR workflow for saffron authentication, covering origin and purity analysis.

The Scientist's Toolkit: Key Research Reagents and Materials

Successful implementation of IR spectroscopy for food analysis relies on a set of essential reagents, materials, and software tools.

Table 2: Essential Research Reagents and Materials

Item/Category	Function/Description	Example Applications
FT-NIR Spectrometer	Instrument for acquiring NIR spectra; often with fiber optic probes for flexible sampling.	Quantitative analysis of protein, moisture, fat in grains, milk [8] [2].
FTIR Spectrometer with ATR	Instrument for acquiring MIR spectra; ATR accessory allows for minimal sample prep.	Detection of adulteration in oils, honey, dairy; saffron authentication [4] [6].
Vibrational Probes (e.g., Azide, ¹³C, Deuterium)	Bioorthogonal tags used as metabolic probes to track specific pathways in complex systems.	Metabolic imaging in biological samples; tracking anabolism of specific nutrients [9].
Chemometrics Software	Software for spectral pre-processing, multivariate calibration, and classification.	Developing PLSR models for quantification; PLS-DA for classification [7] [2].
Reference Materials	Certified materials for instrument validation and calibration model development.	Ensuring accuracy of quantitative models for protein, fat, etc. [10].
(R)-IPrPhanePHOS	(R)-IPrPhanePHOS	High-purity (R)-IPrPhanePHOS, a chiral bisphosphine ligand. For Research Use Only. Not for human or veterinary diagnosis or therapeutic use.
13-Docosen-1-ol, (13Z)-	13-Docosen-1-ol, (13Z)-, MF:C22H44O, MW:324.6 g/mol	Chemical Reagent

Diagram 2: Universal chemometric workflow for processing NIR and MIR spectral data.

The core principles of molecular vibrations provide the foundation for understanding and applying NIR and MIR spectroscopy in food analysis. While MIR spectroscopy probes fundamental vibrations, offering detailed molecular fingerprints, NIR spectroscopy utilizes overtones and combinations for rapid, deep-penetration analysis of bulk constituents. The successful application of both techniques is inextricably linked to robust chemometric methods for spectral pre-processing and multivariate modeling. The presented experimental protocols and toolkit provide a practical framework for researchers to implement these powerful techniques. As the field advances, the integration of more sophisticated machine learning models and high-throughput imaging promises to further unlock the potential of infrared spectroscopy, solidifying its role as a green and efficient solution for ensuring food quality and authenticity.

The analysis of food quality and authenticity is a critical challenge in ensuring consumer safety and compliance with global regulatory standards. Infrared spectroscopy has emerged as a cornerstone analytical technique in this field, providing rapid, non-destructive assessment of food matrices. A significant evolution in this domain is the transition from traditional, laboratory-bound benchtop instruments to agile, portable spectrometers that enable analysis directly in the field and throughout the supply chain [11] [12]. This overview details the principles, capabilities, and applications of both benchtop Fourier-Transform Infrared (FTIR) and portable Near-Infrared (NIR) spectrometers, providing a structured comparison and detailed experimental protocols for their application in food research.

Technical Foundations and Key Comparisons

Vibrational spectroscopy, including FTIR and NIR, operates on the principle of measuring the interaction between infrared light and matter. When molecules are exposed to infrared radiation, they absorb specific wavelengths that correspond to the energies of their chemical bonds' vibrational modes. This produces a spectral "fingerprint" unique to the sample's chemical composition.

Benchtop FTIR spectrometers typically utilize an interferometer and operate across the mid-infrared region (MIR, approximately 4000-400 cm⁻¹), which is rich in fundamental vibrational transitions and allows for highly specific compound identification [11] [13].

Portable NIR spectrometers operate in the near-infrared region (approximately 780-2500 nm), which encompasses overtone and combination bands of C-H, N-H, and O-H bonds. While these signals are broader and more complex to interpret, they are well-suited for quantitative analysis and require advanced chemometrics for deconvolution [14] [15]. The defining characteristic of portable NIR devices is their miniaturization, achieved through advancements in microelectro-mechanical systems (MEMS) and microelectronics, enabling their use outside traditional laboratories [12].

The table below summarizes the core characteristics of these two instrument classes.

Table 1: Comparative Analysis of Benchtop FTIR and Portable NIR Spectrometers

Characteristic	Benchtop FTIR Spectrometer	Portable NIR Spectrometer
Typical Spectral Range	Mid-IR (4000 - 400 cm⁻¹) [13]	Near-IR (780 - 2500 nm) [14] [16]
Primary Analytical Strength	High specificity for compound identification [11]	Rapid quantification and classification [11] [15]
Throughput & Destructiveness	High-throughput; typically non-destructive [11]	Rapid; non-destructive [11] [14]
Portability & Use Case	Laboratory-bound; controlled environments	Handheld; on-site at farm, processing line, or market [15] [12]
Sample Preparation	Often required	Minimal to none [17] [16]
Spectral Resolution	High	Generally lower than benchtop systems [12]
Key Application Example	Detection of specific adulterants in oils [11]	Screening for pesticide residues on intact fruits [14] [16]

Experimental Protocols

Protocol 1: Detection of Pesticide Residues on Intact Fruits using Portable NIR

This protocol is adapted from a study demonstrating the quantification of pesticides like azoxystrobin and chlorpyrifos in cherry tomatoes and strawberries [14].

1. Sample Preparation:

• Acquire intact fruits (e.g., strawberries, tomatoes). For method development, use organically grown fruits to ensure no pre-existing pesticide residues.
• For calibration, prepare samples by spraying with pesticide solutions at different concentrations, reflecting the maximum residue limits (MRLs) and typical use patterns. Include a set of untreated control samples.
• Allow the sprayed samples to dry at room temperature before spectral acquisition.

2. Instrumentation and Spectral Acquisition:

• Device: Use a portable NIR spectrometer with a wavelength range of at least 900-1700 nm [14].
• Setup: Operate the device in reflectance mode. Ensure the device is calibrated according to the manufacturer's instructions.
• Measurement: Take multiple spectra from different positions on each fruit's surface to account for natural variability. A minimum of three scans per fruit is recommended. Ensure consistent contact or distance between the spectrometer's probe and the fruit surface.

3. Data Preprocessing:

• Convert raw reflectance spectra to absorbance (A = log(1/R)).
• Apply preprocessing techniques to minimize light scattering effects and enhance spectral features. Common methods include:
  • Standard Normal Variate (SNV)
  • Multiplicative Scatter Correction (MSC)
  • Derivatives (e.g., first or second derivative) to resolve overlapping peaks and remove baseline drift [14] [17].

4. Chemometric Modeling and Validation:

• Model Development: Use the preprocessed spectra to build a quantification model.
  • Recommended Algorithm: Orthogonal Projections to Latent Structures (OPLS) regression combined with variable selection methods (e.g., Recursive Feature Elimination) has shown high predictive capacity (R²cv > 0.9) for this application [14].
• Validation: Split the dataset into a calibration (training) set and a validation (test) set. Validate model performance using cross-validation and external validation sets to report key metrics such as Root Mean Square Error of Cross-Validation (RMSECV) and the coefficient of determination for the prediction set (R²p) [14].

The following workflow diagram illustrates the key steps of this protocol.

Protocol 2: Authenticity Assessment and Adulteration Detection using Benchtop FT-NIR

This protocol is based on studies comparing benchtop and portable systems for detecting citric acid adulteration in lime juice [17].

1. Sample Preparation:

• Prepare genuine juice samples using a cold press juicer.
• Homogenize the juices thoroughly using an ultra-turrax homogenizer to ensure spectral consistency.
• Prepare adulterated samples by adding exogenous citric acid to genuine juice at varying levels.
• For benchtop analysis, transfer the liquid sample into a standardized quartz cuvette with a fixed path length (e.g., 2 mm) [17].

2. Instrumentation and Spectral Acquisition:

• Device: Use a benchtop FT-NIR spectrometer, typically covering 1000-2500 nm [17].
• Setup: Calibrate the instrument using a built-in external reference (e.g., a background air scan or a certified reference standard) before each measurement session.
• Measurement: Acquire triplicate spectra for each sample in transmittance or diffuse reflectance mode, as appropriate. The high stability of the benchtop system allows for high-resolution scanning at a controlled temperature.

3. Data Preprocessing:

• Similar to Protocol 1, apply necessary preprocessing. For authenticity classification, Standard Normal Variate (SNV) or Multiplicative Scatter Correction (MSC) are highly effective for benchtop FT-NIR data [17].

4. Chemometric Modeling and Validation:

• Exploratory Analysis: Begin with Principal Component Analysis (PCA) to observe natural clustering and identify outliers.
• Discriminant Model: Use Partial Least Squares Discriminant Analysis (PLS-DA) to build a model that classifies samples as "genuine" or "adulterated." This model can achieve high accuracy (>94%) [17].
• Class-Modeling Approach: For authenticity problems, Soft Independent Modeling of Class Analogy (SIMCA) is particularly powerful. It creates a model for the "genuine" class, and any sample that does not fit this model is flagged as atypical or adulterated. This approach has shown overall model performance of up to 98% for benchtop systems [17].

The Scientist's Toolkit: Essential Research Reagents and Materials

The table below lists key materials and software solutions essential for conducting research in this field.

Table 2: Essential Research Reagents and Materials for Spectroscopic Food Analysis

Item	Function/Application
Portable NIR Spectrometer (e.g., viavi MicroNIR, Thermo Fisher Phazir)	Handheld device for on-site, non-destructive spectral data collection on intact samples [12].
Benchtop FT-NIR Spectrometer (e.g., Buchi N-500)	High-performance laboratory instrument for high-resolution spectral analysis of liquid and solid samples [17].
Chemometrics Software (e.g., PLS_Toolbox, The Unscrambler)	Software for advanced multivariate data analysis, including PCA, PLS, OPLS, and machine learning algorithms [11] [12].
Reference Analytical Standard (e.g., Certified Citric Acid)	Used for preparing calibration samples with known adulterant concentrations to build and validate predictive models [17].
Ultra-Turrax Homogenizer	Ensures sample homogeneity, which is critical for obtaining reproducible and reliable spectra, especially for liquid and semi-solid matrices [17].
Standardized Quartz Cuvettes	Provides a consistent and reproducible path length for liquid sample analysis in benchtop spectrometers [17].
1-Phenylpiperazin-2-imine	1-Phenylpiperazin-2-imine CAS 693210-97-8 - For Research
zinc;methoxybenzene	zinc;methoxybenzene, CAS:684215-27-8, MF:C14H14O2Zn, MW:279.6 g/mol

Data Analysis and Chemometric Workflow

The powerful synergy between spectroscopy and chemometrics is what enables the extraction of meaningful information from complex spectral data. The process involves a logical sequence of steps, from raw data to actionable results, as illustrated below.

Key Chemometric Techniques:

• Data Preprocessing: Techniques like SNV and derivatives are crucial for removing physical light-scattering effects and enhancing chemical-related spectral features prior to modeling [14] [17].
• Exploratory Analysis (PCA): An unsupervised method used to visualize inherent data structure, identify patterns, clusters, and potential outliers without prior knowledge of sample classes [17] [12].
• Predictive Modeling: This includes both discriminant and class-modeling techniques.
  • PLS-DA and OPLS: Supervised methods that maximize the separation between predefined sample classes (e.g., contaminated vs. clean). OPLS, in particular, separates the variation related to the predictive component from orthogonal (unrelated) variation, often leading to more interpretable models [14].
  • SIMCA: A class-modeling technique that defines the boundaries of a target class (e.g., "authentic" honey). It is highly suited for authenticity testing as it can identify samples that do not belong to the target class, even if they are a new type of adulterant not seen before [17].
• Emerging Trends: The field is rapidly adopting data fusion (integrating data from multiple spectroscopic sources) and deep learning to model complex, non-linear relationships in data, further improving predictive accuracy and robustness [12].

The journey from benchtop FTIR to portable NIR spectrometers marks a paradigm shift in food quality and authenticity testing. Benchtop systems remain the gold standard for high-specificity identification in controlled laboratories, while portable NIR devices empower stakeholders with rapid, on-site screening capabilities. The effective application of both technologies is fundamentally dependent on robust chemometric models. Future directions point toward greater integration of these instruments with the Internet of Things (IoT), cloud computing, and advanced machine learning, paving the way for fully automated, intelligent decision-support systems that ensure food safety and integrity from farm to fork [15] [12].

Infrared (IR) spectroscopy has emerged as a cornerstone analytical technique for addressing critical challenges in food science, namely the assessment of authenticity, detection of adulteration, and evaluation of quality parameters. This family of techniques, which includes Near-Infrared (NIR) and Mid-Infrared (MIR) spectroscopy, along with Fourier-Transform IR (FT-IR) spectroscopy, is prized for its rapid, non-destructive, and green analytical capabilities [2] [18]. Its application is pervasive across the food industry and research sectors, enabling the swift monitoring of chemical composition without extensive sample preparation or the use of hazardous chemicals [19] [20]. When coupled with chemometrics, IR spectroscopy provides a powerful tool for the quantitative analysis of major constituents and the qualitative discrimination of food products based on their unique spectral fingerprints [2] [21]. These applications are vital for ensuring compliance with labeling regulations, protecting consumers from fraudulent practices, and maintaining brand integrity [22] [20]. This document outlines the key applications and provides detailed experimental protocols for implementing these techniques within a research context.

The following table summarizes the primary IR spectroscopy techniques used in food analysis, their principles, and their main applications in quality, authenticity, and adulteration testing.

Table 1: Overview of Infrared Spectroscopy Techniques in Food Analysis

Technique	Spectral Range	Principle of Operation	Strengths	Common Food Applications
Near-Infrared (NIR) Spectroscopy [2] [18]	750–2500 nm (12,500–4000 cm⁻¹)	Measures overtones and combinations of fundamental vibrations from C-H, N-H, and O-H bonds.	Rapid, high penetration depth, minimal sample preparation, suitable for online/at-line analysis.	Quantification of protein, moisture, fat, carbohydrates in grains, meat, dairy [2] [19]; Identification of origin; Adulteration screening.
Fourier-Transform Mid-Infrared (FT-IR) Spectroscopy [23] [24]	4000–400 cm⁻¹	Measures fundamental molecular vibrations, providing detailed chemical structure information.	High specificity and information content; excellent for identifying functional groups and specific compounds.	Authentication of edible oils [23]; Detection of adulteration in honey, spices, milk [25] [22]; Analysis of physicochemical properties.
Raman Spectroscopy [25]	Varies (Laser-dependent)	Measures inelastic scattering of light, providing a vibrational fingerprint of the sample.	Minimal interference from water; provides complementary information to IR.	Detection of toxic substances, foodborne pathogens, and alcohol content in beverages [25].

Key Applications and Quantitative Data

The application of IR spectroscopy spans a vast array of food matrices. The table below compiles specific use cases and, where available, quantitative performance data from recent research.

Table 2: Key Applications and Performance of IR Spectroscopy in Food Testing

Food Category	Analyte/Parameter of Interest	Technique Used	Key Findings & Performance Metrics
Coffee [25]	Trace Elements (As, Pb, Cr, Zn, Fe, etc.)	ICP-OES	Successfully quantified 10 trace elements. Highest average concentration was Fe (498.72 ± 23.07 μg/kg). All samples were within safe limits.
Chicken Meat [25]	Geographical Origin	ICP-Based Methods	OPLS-DA model identified 23-28 significant elements for discrimination. Canonical discriminant analysis achieved 100% accuracy in classification.
Edible Oils [23] [22]	Authenticity & Adulteration	FT-IR	Successfully used to detect and quantify adulteration with lower-grade oils. Coupled with chemometrics for rapid, multi-parameter prediction.
Beverages & Spirits [25]	Ethanol & Toxic Alcohols (Methanol)	Raman Spectroscopy	Enabled rapid, non-destructive measurement through container. Applied for health safety and identifying adulteration.
Milk & Dairy [19] [26]	Fat, Protein, Lactose, Adulteration	FT-NIR / IR	Standard for rapid quantification of major components. Used for raw milk authentication by comparing spectral fingerprints to a known pure reference.
Grains & Flour [2] [20]	Protein, Moisture, Starch, Fiber	NIRS	Routine quality control. Diffuse reflectance mode used for solids/powders. Provides results in seconds for multiple parameters.
Fruits & Vegetables [18] [20]	Sugar (Brix), Ripeness, Internal Defects	NIRS (Interactance)	Non-destructive assessment of internal quality. Suitable for whole, intact produce.
Plastic Food Packaging [25]	Heavy Metals (Co, As, Cd, Pb, etc.)	ICP-MS	Method validated with LOD: 0.10–0.85 ng/mL; LOQ: 0.33–2.81 ng/mL. Migration of Zn, Al, and Pb into foodstuffs was confirmed.

Experimental Protocols

Protocol 1: FT-IR Analysis for Edible Oil Authenticity

Application: Detection of adulteration in extra virgin olive oil [23] [22].

Principle: Adulterants (e.g., cheaper vegetable oils) introduce distinct chemical functional groups or alter the ratios of existing ones (e.g., C=O, C-H stretches), leading to detectable changes in the FT-IR spectrum.

Materials & Reagents:

• Pure, authenticated extra virgin olive oil reference samples.
• Test oil samples.
• FT-IR spectrometer with ATR (Attenuated Total Reflectance) accessory.
• Liquid sample cell or ATR crystal (e.g., diamond, ZnSe).
• Solvent (e.g., hexane) for cleaning; Lint-free wipes.

Procedure:

• Instrument Start-up: Power on the FT-IR spectrometer and allow it to stabilize for at least 15 minutes.
• Background Collection: Clean the ATR crystal thoroughly with solvent and dry. Collect a background spectrum of the clean crystal.
• Sample Loading: Apply a small drop of the pure reference oil or test oil directly onto the ATR crystal, ensuring full coverage of the crystal surface.
• Spectral Acquisition:
  • Acquire spectra in the range of 4000–400 cm⁻¹.
  • Set resolution to 4 cm⁻¹ and accumulate 32–64 scans per spectrum to ensure a high signal-to-noise ratio.
• Replication: Analyze each sample in at least triplicate, cleaning the crystal between each measurement.
• Data Pre-processing: Export the averaged spectra and apply necessary pre-processing steps such as Standard Normal Variate (SNV) or derivative filters (e.g., Savitzky–Golay) to minimize baseline drift and scattering effects [2].

Data Analysis & Chemometrics:

• Qualitative Analysis (PCA): Use Principal Component Analysis (PCA) on the pre-processed spectral data to observe natural clustering. Adulterated samples will appear as outliers or form separate clusters from the pure reference samples.
• Quantitative Analysis (PLSR): To quantify the level of adulteration, use Partial Least Squares Regression (PLSR). Develop a calibration model by spiking pure oil with known concentrations of an adulterant and recording their spectra. The model correlates spectral changes with adulterant concentration, allowing for prediction in unknown samples [23].

Protocol 2: NIRS for Quantitative Analysis of Solid Food Powders

Application: Determination of protein, moisture, and fat in milk powder [19] [20].

Principle: Chemical bonds (O-H, N-H, C-H) in major food components absorb NIR light at specific wavelengths. The intensity of absorption is related to their concentration.

Materials & Reagents:

• Milk powder samples.
• NIR spectrometer equipped with a diffuse reflectance cup or spinning module.
• Standard laboratory reference methods (e.g., Kjeldahl for protein, loss on drying for moisture).
• Cuvettes or sample cups for powder analysis.

Procedure:

• Sample Preparation: Ensure samples are homogeneous. For powders, consistent particle size is critical; grinding may be necessary. Allow samples to equilibrate to room temperature.
• Instrument Calibration: This is a critical one-time/periodic step. Analyze a large set (n > 50) of representative samples using both the NIR spectrometer and the standard reference methods.
• Spectral Acquisition:
  • Fill the sample cup consistently and evenly without compacting.
  • Acquire spectra in the 800–2500 nm range.
  • Use a rotating cup or collect multiple spectra from different spots to account for heterogeneity.
• Model Development: Use chemometric software to build a calibration model (typically using PLSR) that correlates the spectral data (X-matrix) with the reference analytical data (Y-matrix). Pre-process spectra using Multiplicative Scatter Correction (MSC) or derivatives as needed [2].
• Validation: Validate the model using an independent set of samples not used in the calibration. Key performance metrics include Coefficient of Determination (R²), Standard Error of Prediction (SEP), and Residual Predictive Deviation (RPD).

Data Analysis: Once a robust calibration model is established and loaded into the instrument software, routine analysis involves simply acquiring the spectrum of an unknown sample, and the software instantly predicts the values for protein, moisture, fat, etc., based on the model.

Workflow Diagram: IR-Based Food Analysis

The following diagram illustrates the generalized workflow for authenticity and quality control using IR spectroscopy.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Reagents for IR-based Food Analysis

Item / Solution	Function / Purpose	Application Example / Notes
FT-IR Spectrometer with ATR	Enables direct analysis of liquids, pastes, and solids with minimal preparation.	Authentication of oils, honey, and liquid dairy products [23] [22].
NIR Spectrometer with Reflectance Probe	For non-destructive analysis of solid and powdered samples in lab or inline.	Analysis of grain, flour, and powdered milk in a production environment [19] [20].
Chemometric Software Package	Critical for developing quantitative (PLSR) and qualitative (PCA, SIMCA) models from spectral data.	Open-source (R, Python) or commercial packages (OPUS, Unscrambler) are used [2] [18].
Certified Reference Materials (CRMs)	Used for instrument verification and as primary standards for calibration models.	NIST-traceable standards for protein, moisture, etc., to ensure model accuracy [25].
Microfluidic Chips & SERS Substrates	Used with Raman spectroscopy to trap and enhance signal from specific analytes like pathogens.	Detection of foodborne pathogens; requires specific substrate functionalization [25].
Solvents (Hexane, Ethanol)	For cleaning optics and ATR crystals between samples to prevent cross-contamination.	High-purity, spectroscopic grade is recommended to avoid introducing spectral artifacts.
3-Bromocyclohept-2-en-1-one	3-Bromocyclohept-2-en-1-one	3-Bromocyclohept-2-en-1-one (C7H9BrO) is a brominated cyclic enone for synthetic organic chemistry research. This product is for Research Use Only (RUO). Not for human or veterinary use.
4-Bromohex-4-en-3-one	4-Bromohex-4-en-3-one|CAS 811470-76-5|RUO	High-purity 4-Bromohex-4-en-3-one (C6H9BrO) for research. An α,β-unsaturated carbonyl building block for synthetic chemistry. For Research Use Only. Not for human or veterinary use.

From Theory to Practice: Methodological Applications Across Food Matrices

Infrared spectroscopy, particularly in the near-infrared (NIR) region, has established itself as a cornerstone technique for rapid, non-destructive analysis in food quality and authenticity testing [3] [27]. The spectral band of NIR typically covers 780 nm to 2500 nm, measuring the interaction between NIR radiation and chemical bonds in the sample, primarily those in hydrogen-containing groups (O-H, C-H, N-H) [3]. However, the resulting spectra are complex, characterized by broad, overlapping absorption bands that are difficult to interpret directly [3] [28]. This is where chemometrics transforms spectral data into actionable insight.

Chemometrics, defined as the mathematical extraction of relevant chemical information from measured analytical data, integrates theories from mathematics, statistics, and computer science to identify, quantify, and classify sample properties [3] [28]. The integration of artificial intelligence (AI) and machine learning (ML) with classical chemometric methods represents a paradigm shift, enabling the analysis of complex, high-dimensional datasets that overwhelm traditional techniques [12] [29] [28]. This document provides application notes and detailed protocols for utilizing Principal Component Analysis (PCA), Partial Least Squares (PLS) regression, and machine learning for spectral analysis within food research.

Theoretical Foundations of Chemometric Techniques

Principal Component Analysis (PCA)

PCA is an unsupervised learning technique used for exploratory data analysis and dimensionality reduction [28]. It projects the original, potentially correlated spectral variables into a new set of orthogonal variables called Principal Components (PCs). The first PC captures the maximum variance in the data, with each subsequent component capturing the remaining variance in descending order. This allows for the visualization of sample clustering, identification of outliers, and detection of natural patterns within high-dimensional spectral data without prior knowledge of sample classes [28].

Partial Least Squares (PLS) Regression

PLS regression is a supervised multivariate calibration method used to model the relationship between a spectral matrix (X) and a vector of concentration values or reference analyses (Y) [3] [30]. Unlike PCA, which only considers the variance in X, PLS finds latent variables that simultaneously maximize the covariance between X and Y. This makes it particularly powerful for predicting analyte concentrations in complex mixtures like foodstuffs, even in the presence of collinearity and noise [30] [31]. Key metrics for evaluating PLS models include the Coefficient of Determination (R²) and the Root Mean Square Error of Calibration (RMSEC) or Prediction (RMSEP) [31].

Machine Learning (ML) in Spectroscopy

Machine learning algorithms significantly expand the toolbox available for spectral analysis. They are particularly adept at handling non-linear relationships and automating feature extraction [29] [28].

• Support Vector Machine (SVM): A supervised algorithm that finds the optimal hyperplane to separate classes or predict quantitative values in high-dimensional space. It is effective for classification and regression tasks, especially with limited samples [28].
• Random Forest (RF): An ensemble method that constructs multiple decision trees and aggregates their results. It offers strong generalization, reduces overfitting, and provides feature importance rankings [29] [28].
• Convolutional Neural Networks (CNN): A class of deep learning models capable of automatically learning hierarchical features from raw or minimally preprocessed spectral data, often achieving state-of-the-art performance in classification and quantification tasks [29] [28].

Application Notes: Chemometrics in Food Quality and Authenticity

The following table summarizes selected recent applications of chemometric techniques in food analysis, demonstrating their performance across various matrices and challenges.

Table 1: Applications of Chemometric Techniques in Food Analysis

Food Matrix	Analytical Challenge	Chemometric Technique(s)	Key Performance Metrics	Reference Source
Tiger Nut (Cyperus esculentus L.)	Quantification of crude oil, protein, and starch	PLSR with variable selection	R²: 0.8946 (oil), 0.8525 (protein), 0.8778 (starch)	[31]
Honey	Authentication & detection of adulteration	PCA, PLSR, Linear Discriminant Analysis (LDA)	Classification accuracy >90% for adulteration	[30]
Plant-Based Milk Alternatives	Classification and detection of variability	PCA, Hierarchical Cluster Analysis (HCA)	Successful discrimination of almond, oat, rice, and soy drinks	[32]
Milk	Geographical origin traceability	Portable NIR with Fuzzy Direct LDA-KNN	Classification accuracy of 97.33%	[3]
Peanut Oil	Identification of adulteration	PLS Modeling	R² > 0.9311, low RMSECV	[3]
Powdered Foods (spices, dairy, cereals)	Adulterant detection	SVM, Random Forest, Deep Learning	Detection accuracy often >90%	[33]

Workflow for Spectral Analysis

A robust chemometric analysis follows a structured workflow from spectral acquisition to model deployment. The following diagram illustrates the key stages, highlighting the iterative nature of model development and validation.

Detailed Experimental Protocols

Protocol 1: Quantification of Nutrient Composition in Tiger Nuts using PLSR

Objective: To rapidly and non-destructively predict the content of crude oil (CO), crude protein (CP), and total starch (TS) in tiger nut tubers using a portable NIR spectrometer and PLSR [31].

Materials and Reagents: Table 2: Research Reagent Solutions & Essential Materials

Item	Function / Description	Example / Specification
Portable NIR Spectrometer	Acquisition of spectral data from samples.	IAS8120 (Range: 800-1100 nm or broader) [31]
Tiger Nut Samples	Representative sample set for calibration and validation.	75 samples, 28 varieties, multiple regions [31]
Reference Chemistry: Soxhlet Apparatus	Determination of reference crude oil content for model calibration.	Standard solvent extraction [31]
Reference Chemistry: Kjeldahl Apparatus	Determination of reference crude protein content for model calibration.	Measures nitrogen content [31]
Reference Chemistry: Spectrophotometer	Determination of reference total starch content for model calibration.	Dual-wavelength colorimetric method [31]
Data Analysis Software	For spectral preprocessing, variable selection, and PLSR modeling.	Python, R, MATLAB, or commercial chemometrics software

Procedure:

• Sample Preparation: Wash fresh tiger nut tubers and air-dry to a stable moisture content (~10%). Grind the dried tubers and sieve through a 30-mesh screen. Store prepared powder at 4°C [31].
• Reference Analysis: Determine the CO, CP, and TS content for each sample using standard chemical methods (Soxhlet extraction, Kjeldahl, and spectrophotometry, respectively). These values form the reference (Y-block) for the PLSR model [31].
• Spectral Acquisition: Using the portable NIR spectrometer, collect the spectra of all powdered samples. Ensure consistent scanning parameters and environmental conditions. The resulting spectra form the predictor matrix (X-block) [31].
• Spectral Preprocessing: Apply preprocessing techniques to minimize light scattering and noise. Common methods include:
  • Savitzky-Golay (SG) Smoothing: Reduces high-frequency noise [33].
  • Standard Normal Variate (SNV): Corrects for scatter effects due to particle size differences [33].
• Variable Selection (Optional but Recommended): To improve model performance, employ algorithms like Interval Partial Least Squares (iPLS) or Moving Window PLS (MWPLS) to select spectral regions most informative for the specific component being analyzed [31].
• Model Development & Validation:
  • Split the dataset into a calibration set (e.g., 70-80% of samples) and a validation set (20-30%).
  • Build the PLSR model on the calibration set, correlating the preprocessed spectra (X) with the reference values (Y).
  • Use leave-one-out or cross-validation on the calibration set to determine the optimal number of latent variables and prevent overfitting.
  • Validate the model by predicting the component values in the independent validation set. Calculate performance metrics: R² and RMSEP [31].

Protocol 2: Authentication of Honey and Detection of Adulteration

Objective: To use NIR spectroscopy combined with chemometrics to verify honey authenticity, classify botanical origin, and detect adulterants like corn or rice syrup [30].

Materials and Reagents:

• NIR Spectrometer (benchtop or portable with InGaAs detector)
• Pure and Adulterated Honey Samples
• Liquid Sample Cell or Transflectance Probe
• Chemometrics Software with PCA, PLS-DA, and LDA algorithms.

Procedure:

• Sample Preparation: Ensure honey samples are well-mixed and free of air bubbles. Warm crystallized samples gently in a water bath (not exceeding 40°C) to liquefy, then homogenize. Allow samples to equilibrate to room temperature (e.g., 25°C) before analysis [30].
• Spectral Acquisition: Collect NIR spectra of all honey samples using a transmission or transflectance cell. A wavelength range of 1000-2500 nm is typical. For each sample, collect multiple scans and average them to improve the signal-to-noise ratio [30].
• Data Preprocessing: Apply standard preprocessing techniques to the raw spectra. Common choices for honey include:
  • Multiplicative Scatter Correction (MSC) or SNV to compensate for path length differences and scattering.
  • Savitzky-Golay First or Second Derivative to enhance spectral features and remove baseline offsets [30].
• Exploratory Analysis (PCA): Perform PCA on the preprocessed spectral data. Examine the scores plot (e.g., PC1 vs. PC2) to visualize natural clustering of samples, identify potential outliers, and observe any separation between pure and adulterated samples or between different floral origins [30].
• Classification Model:
  • For Adulteration Detection: Use a classification algorithm like Linear Discriminant Analysis (LDA) on the principal components from PCA to build a model that classifies samples as "pure" or "adulterated" [30]. Alternatively, use PLS-Discriminant Analysis (PLS-DA) or Support Vector Machine (SVM) directly on the spectra.
  • For Quantifying Adulteration: If the goal is to predict the level of an adulterant, develop a PLSR model using reference values for the adulterant concentration.
• Model Validation: Validate the classification model using an external test set or cross-validation. Report classification accuracy, sensitivity, and specificity. For quantitative PLSR models, report R² and RMSEP [30].

The Integration of Artificial Intelligence and Future Directions

The field is rapidly evolving with the integration of advanced machine learning and AI, moving beyond classical chemometrics. The following diagram illustrates how AI is being integrated into modern spectroscopic workflows.

Key emerging trends include:

• Explainable AI (XAI): Addressing the "black box" nature of complex models like deep learning by providing insights into which spectral regions drive predictions, building trust and providing chemical insight [29] [28].
• Data Fusion: Combining data from multiple spectroscopic sensors (e.g., NIR, Raman, MIR) using low-, mid-, or high-level fusion strategies to create more robust and informative models [12].
• Miniaturization and Portability: The development of handheld and portable spectrometers, coupled with robust chemometric models, enables real-time, on-site analysis throughout the food supply chain [12] [27].
• Self-Adaptive Models and Multi-Omics Integration: Future frameworks may involve models that continuously learn from new data. Furthermore, using AI to fuse spectral data with other 'omics' data (genomics, proteomics) promises a more holistic understanding of food quality and authenticity [29] [33].

In the rapidly expanding plant-based food sector, product authentication and quality control have become paramount for consumer protection and regulatory compliance. This case study details the application of Attenuated Total Reflectance Fourier Transform Infrared (ATR-FTIR) spectroscopy for authenticating commercial plant-based milk substitutes. Framed within broader thesis research on infrared spectroscopy for food quality and authenticity testing, this work demonstrates how ATR-FTIR, combined with chemometric analysis, provides a rapid, non-destructive method for classifying plant-based beverages and detecting potential adulteration or compositional variability.

The global market for plant-based milk alternatives has experienced remarkable growth, with per capita revenue expected to increase by approximately 127% from 2014 to 2027 [32]. This surge, driven by environmental concerns, health considerations, and dietary preferences, has created an urgent need for analytical methods to verify product authenticity and compositional integrity [32]. Plant-based beverages derived from almonds, oats, rice, and soy exhibit distinct biochemical profiles that serve as chemical fingerprints for authentication purposes [34].

Theoretical Background

ATR-FTIR Spectroscopy Fundamentals

ATR-FTIR spectroscopy leverages the interaction between infrared radiation and molecular bonds in a sample to produce characteristic spectral fingerprints. When IR light interacts with a sample placed on a crystal with a high refractive index, it undergoes total internal reflection, generating an evanescent wave that penetrates the sample typically to a depth of 0.5-5 micrometers. Molecular bonds within the sample absorb specific frequencies of this radiation, resulting in absorption bands that correspond to the sample's chemical composition [35].

The resulting spectrum provides a comprehensive molecular fingerprint of the sample, with absorption bands representing specific molecular vibrations from functional groups present in proteins, carbohydrates, lipids, and other constituents. For plant-based milk authentication, the Amide I and II regions (1700-1600 cm⁻¹ and 1600-1500 cm⁻¹) are particularly important for protein characterization, while the carbohydrate region (1200-900 cm⁻¹) provides information about sugar and fiber content [32] [34].

Chemometrics in Spectral Analysis

Chemometrics applies statistical and mathematical methods to extract meaningful information from chemical data. When applied to complex ATR-FTIR spectral data, chemometric techniques enable:

• Pattern recognition for sample classification
• Discrimination between categories based on compositional differences
• Quantification of specific components in complex mixtures

Principal Component Analysis (PCA) reduces the dimensionality of spectral data while preserving variance, allowing visualization of natural clustering between sample classes. Hierarchical Cluster Analysis (HCA) groups samples based on spectral similarity, while Partial Least Squares Discriminant Analysis (PLS-DA) and Orthogonal PLS-DA (OPLS-DA) build predictive models for classification [32] [34].

Experimental Protocols

Sample Preparation

Table 1: Sample Preparation Protocol for Plant-Based Milk Analysis Using ATR-FTIR

Step	Procedure	Parameters	Purpose
Sample Acquisition	Purchase commercial plant-based beverages from retail markets	Include multiple brands and batches for each beverage type (almond, oat, rice, soy)	Ensure representative sampling of commercial products
Lyophilization	Freeze samples at -80°C for 3 days, then lyophilize	Vacuum: 0.80 mbar; Temperature: -25°C until completely dehydrated [34]	Remove water interference from spectra; preserve macronutrient structure
Homogenization	Gently mix or vortex samples before analysis	Ensure uniform distribution of components	Improve spectral reproducibility

Sample preparation is critical for obtaining high-quality, reproducible ATR-FTIR spectra. Lyophilization eliminates the strong infrared absorption of water, which can obscure important spectral regions, particularly the Amide I region crucial for protein secondary structure analysis [34]. The freeze-drying process preserves the native structure of macromolecules, ensuring that spectral features accurately represent the beverage's composition.

ATR-FTIR Spectral Acquisition

Table 2: Instrumental Parameters for ATR-FTIR Spectral Acquisition

Parameter	Specification	Rationale
Instrument	FTIR Spectrometer with ATR accessory	Standard equipment for solid and liquid sample analysis
ATR Crystal	Diamond internal reflection element	Durability and chemical resistance; suitable for diverse samples
Spectral Range	4000-400 cm⁻¹	Covers fundamental molecular vibration regions
Resolution	4 cm⁻¹	Optimal balance between spectral detail and signal-to-noise ratio
Scan Number	20-100 scans per sample	Improve signal-to-noise ratio through averaging
Background Correction	Before each sample or sample set	Minimize atmospheric interference (COâ‚‚, Hâ‚‚O vapor)

Spectral acquisition follows a standardized protocol: (1) Clean the ATR crystal with appropriate solvents and dry; (2) Collect background spectrum; (3) Apply lyophilized sample to completely cover the crystal surface; (4) Apply consistent pressure using the instrument's pressure clamp; (5) Collect sample spectra; (6) Clean crystal thoroughly between samples [34]. Most studies employ 20-100 scans per sample to ensure adequate signal-to-noise ratio while maintaining practical analysis time.

Data Pre-processing and Chemometric Analysis

Raw spectral data requires pre-processing to remove instrumental artifacts and enhance chemical information before chemometric analysis:

• ATR Correction: Compensates for depth of penetration variation with wavelength
• Baseline Correction: Removes scattering effects and baseline drift
• Normalization: Standardizes spectral intensity for comparative analysis (typically Min-Max or Standard Normal Variate)
• Smoothing: Reduces high-frequency noise (Savitzky-Golay filter commonly used)
• Derivative Analysis: Second-derivative transformation enhances resolution of overlapping bands, particularly in the Amide I region for protein secondary structure analysis [34]

Following pre-processing, both unsupervised (PCA, HCA) and supervised (PLS-DA, OPLS-DA) chemometric methods are applied to classify samples and identify discriminatory spectral features.

Key Experimental Findings

Spectral Features for Classification

Table 3: Characteristic ATR-FTIR Spectral Regions for Plant-Based Milk Authentication

Spectral Region (cm⁻¹)	Molecular Assignment	Discriminatory Utility	Key Findings
3000-2800	C-H stretching (lipids, fatty acids)	Differentiation based on lipid profiles	Variations in almond beverages due to different lipid content [32]
1700-1600 (Amide I)	C=O stretching (proteins)	Protein secondary structure quantification	β-turn and α-helix structures key for discrimination [34]
1600-1500 (Amide II)	N-H bending, C-N stretching (proteins)	Protein content and characteristics	Soy beverages show distinct protein profile [34]
1200-900	C-O-C, C-O stretching (carbohydrates)	Carbohydrate profile differentiation	Distinguishes oat (β-glucans) and rice (high sugar) beverages [32] [34]

Research demonstrates that ATR-FTIR spectroscopy effectively discriminates between different types of plant-based beverages. In a comprehensive study analyzing 41 commercial beverages, soy and rice beverages formed distinct clusters in chemometric models, while almond and oat samples showed partial overlap due to greater compositional variability [34]. Variable Importance in Projection (VIP) scores from OPLS-DA models identified β-turn and α-helix protein structures, along with carbohydrate-associated spectral bands, as the most significant features for classification [34].

Challenges in Almond Beverage Authentication

Almond-based beverages present particular challenges for authentication due to their significant compositional variability. ATR-FTIR studies have revealed that almond beverages often demonstrate less precise clustering in chemometric models compared to oat, rice, and soy beverages [32]. This variability frequently stems from the inclusion of other ingredients like rice or soy as fillers or stabilizers, which can mislead consumers about nutritional content [32]. The ATR-FTIR method successfully detects this variability, with models accurately identifying almond beverages containing substantial amounts of rice or soy components.

The Scientist's Toolkit

Table 4: Essential Research Reagent Solutions for ATR-FTIR Analysis of Plant-Based Milks

Item	Specification	Function/Application
FTIR Spectrometer	Equipped with ATR accessory (diamond crystal recommended)	Spectral acquisition from plant-based milk samples
Lyophilizer	Capable of reaching -80°C and 0.80 mbar vacuum	Sample dehydration to remove water interference
Chemometrics Software	MATLAB, Python with scipy.stats, or dedicated spectral analysis packages	Multivariate data analysis and classification model development
Reference Materials	Pure plant sources (almond, oat, rice, soy)	Method validation and calibration standards
Crystal Cleaning Solvents	HPLC-grade solvents (e.g., methanol, ethanol)	ATR crystal cleaning between samples to prevent cross-contamination
8-Chloroisoquinolin-4-ol	8-Chloroisoquinolin-4-ol, MF:C9H6ClNO, MW:179.60 g/mol	Chemical Reagent
(2R)-2-propyloctanamide	(2R)-2-propyloctanamide

Implementation Workflows

Experimental Workflow

Experimental workflow for plant-based milk authentication

Data Analysis Pathway

Data analysis pathway for spectral interpretation

ATR-FTIR spectroscopy combined with chemometric analysis represents a powerful, rapid, and non-destructive approach for authenticating plant-based milk alternatives. The method successfully discriminates between beverage types based on their unique biochemical fingerprints, with particular effectiveness in identifying protein secondary structures and carbohydrate profiles as key discriminatory features.

Implementation of this analytical approach addresses growing concerns about product authenticity in the expanding plant-based food sector, providing manufacturers, regulators, and researchers with a reliable tool for quality control and verification of label claims. The portability of modern ATR-FTIR instruments further enhances their potential application in various settings, from research laboratories to industrial quality control environments [32].

Future developments in spectral database creation, calibration transfer protocols, and advanced machine learning applications will further strengthen the role of ATR-FTIR spectroscopy in ensuring transparency and authenticity throughout the plant-based food supply chain.

Food fraud represents a significant global challenge to food safety, consumer health, and economic stability, with estimated annual economic losses of $40 billion affecting approximately 16,000 tons of food and beverages [33]. Adulteration manifests in three primary dimensions: intentional, accidental, and falsified [33]. Intentional adulteration includes substituting premium ingredients with cheaper alternatives, such as adding ground nutshells to cinnamon or starches to protein supplements [33]. This case study explores the application of Infrared (IR) spectroscopy, specifically Near-Infrared (NIR) spectroscopy, as a rapid, non-destructive analytical tool for detecting adulteration across three vulnerable food categories: spices, nuts, and liquid foods. Framed within broader thesis research on IR spectroscopy for food quality and authenticity testing, this study provides detailed protocols and data interpretation frameworks suitable for researchers, scientists, and quality control professionals engaged in food fraud mitigation.

Theoretical Foundations of NIR Spectroscopy

Near-Infrared spectroscopy operates in the electromagnetic spectrum range of 800–2500 nm (12,500–4000 cm⁻¹), situated between the visible and mid-infrared regions [36]. This technique measures molecular overtone and combination vibrations, primarily involving C-H, O-H, and N-H chemical bonds [2] [30]. These vibrations provide a unique chemical "fingerprint" for each sample, enabling discrimination between authentic and adulterated products [37].

The interaction between NIR light and matter follows the Beer-Lambert law, where absorbance is proportional to both concentration and optical path length [33]. NIR systems typically comprise a radiation source, sample cell, and detector, with measurements acquired through diffuse reflectance for solids (e.g., spices, nut powders) or transmission/transflectance for liquids (e.g., honey, oils) [33] [2]. The technique's significant advantages include minimal sample preparation, rapid analysis (seconds to minutes), and simultaneous multi-component determination without consuming or altering samples [2] [30].

Despite these advantages, NIR spectroscopy faces limitations including low sensitivity for compounds present at concentrations below 1%, susceptibility to environmental factors (e.g., temperature, moisture), and the necessity for robust calibration models using chemometrics [33] [2]. These limitations underscore the critical importance of proper experimental design and model development, as detailed in subsequent sections.

Table 1: Key Characteristics of NIR Spectroscopy for Food Authentication

Characteristic	Description	Implication for Food Authentication
Spectral Range	800–2500 nm	Captures overtone and combination vibrations of organic compounds
Key Molecular Vibrations	C-H, O-H, N-H bonds	Sensitive to major food components (proteins, fats, carbohydrates, water)
Sample Forms	Solids, liquids, powders	Applicable to diverse food matrices without extensive preparation
Analysis Speed	Seconds to minutes	Suitable for high-throughput screening and real-time decision making
Detection Limits	Typically >0.1–1%	Effective for economically-motivated adulteration at commercially relevant levels
Destructive	Non-destructive	Preserves sample for further testing or legal proceedings

Experimental Design and Workflow

A systematic approach to NIR-based adulteration detection ensures reliable, reproducible results. The following workflow diagram illustrates the comprehensive process from sample preparation to final authentication decision:

Sample Preparation Protocol

For Spices and Nut Powders:

• Grinding: Process samples to uniform particle size using a laboratory mill (e.g., 250–500 μm) to minimize light scattering effects [33].
• Moisture Standardization: Condition samples in a controlled environment (relative humidity 40–50%, 25°C) for 24 hours to reduce spectral variability from water content differences [33].
• Packaging: Present samples in uniform containers with consistent geometry and packing density for reflectance measurements [33].

For Liquid Foods (Honey, Oils, Milk):

• Homogenization: Mix samples thoroughly to ensure uniform distribution of components [30].
• Temperature Equilibration: Standardize sample temperature to 25±1°C to minimize spectral effects from temperature variation [30].
• Cell Selection: Use transmission cells with appropriate path lengths (0.5–2 mm) for liquids, or transflectance cells for viscous samples [2].

Instrumentation and Data Acquisition

Equipment Setup:

• NIR Spectrometer: Benchtop (laboratory) or portable (field) devices covering 1000–2500 nm range [36] [37].
• Detector Type: InGaAs for higher wavelength sensitivity (1100–2500 nm), PbS for diffuse reflectance of solids [2].
• Acquisition Mode: Diffuse reflectance for powdered spices/nuts; transmission for clear liquids; transflectance for viscous liquids [33] [2].

Spectral Collection Parameters:

• Spectral Range: 1000–2500 nm for comprehensive chemical information [30].
• Resolution: 4–16 cm⁻¹ for optimal feature detection [30].
• Scan Number: Minimum 32 scans per spectrum to ensure adequate signal-to-noise ratio [36].
• Replication: Analyze at least 5–10 subsamples per specimen to account for heterogeneity [33].

Spectral Preprocessing and Chemometric Analysis

Spectral Preprocessing Techniques

Raw NIR spectra contain both chemical and physical information, necessitating preprocessing to enhance chemical signals while minimizing confounding physical effects (e.g., light scattering, particle size variation). The table below summarizes common preprocessing techniques and their applications:

Table 2: Spectral Preprocessing Techniques for NIR Analysis of Foods

Technique	Primary Function	Typical Applications	Effect on Spectra
Savitzky-Golay (SG) Smoothing	Reduces high-frequency noise	All sample types; essential before derivative processing	Improves signal-to-noise ratio without significant peak distortion [33]
Standard Normal Variate (SNV)	Corrects for scattering effects	Powdered spices, nut flours, uneven surfaces	Removes multiplicative interferences and baseline shifts [33] [2]
Multiplicative Scatter Correction (MSC)	Compensates for additive and multiplicative scattering	Heterogeneous solid samples	Linearizes reflectance spectra, enhancing chemical information [2]
First Derivative (FD)	Eliminates baseline offset and enhances resolution	Overlapping peaks; subtle spectral features	Emphasizes minor spectral variations; requires subsequent smoothing [33] [2]
Second Derivative (SD)	Enhances peak resolution and class discrimination	Complex mixtures with overlapping absorptions	Improves separation of closely spaced peaks; increases noise [33]

Chemometric Modeling Approaches

Chemometrics applies statistical methods to extract meaningful information from chemical data. For NIR-based authentication, both unsupervised and supervised approaches are employed:

Exploratory Analysis (Unsupervised):

• Principal Component Analysis (PCA): Reduces spectral dimensionality while preserving variance, enabling visualization of natural sample clustering and outlier detection [36] [2]. PCA is particularly valuable for initial data exploration to identify patterns and potential adulteration trends without prior class information.

Classification and Regression (Supervised):

• Partial Least Squares-Discriminant Analysis (PLS-DA): A discriminant version of PLS that maximizes separation between predefined classes (e.g., authentic vs. adulterated) [36].
• Soft Independent Modeling of Class Analogy (SIMCA): A class modeling technique that develops a separate PCA model for each class, useful for verifying sample conformity to a target class (e.g., pure spice) [38].
• Partial Least Squares Regression (PLSR): Establishes relationship between spectral data and continuous reference values (e.g., adulteration percentage), enabling quantitative prediction [2] [30].

Model Validation: Robust validation is essential to ensure model reliability. Employ:

• Cross-validation: Internal validation (e.g., venetian blinds, random subsets) to optimize model complexity and prevent overfitting [36].
• External validation: Testing with completely independent sample sets not used in model development [30].
• Performance metrics: For classification: sensitivity, specificity, accuracy; for regression: RMSECV, RMSEP, R² [2].

Application-Specific Protocols and Data Interpretation

Spice Authentication

Spices represent a high-risk category for adulteration due to their high value and complex supply chains. Common adulterants include spent spice material, foreign plant matter, synthetic dyes, and hazardous substances like Sudan dyes [36] [37].

Experimental Protocol for Black Pepper Authentication:

• Reference Samples: Collect 50 authenticated black pepper samples from verified sources.
• Adulterated Samples: Prepare adulterated samples by mixing authentic pepper with pepper husks, starch, or spent pepper in 5–40% (w/w) increments.
• Spectral Acquisition: Use diffuse reflectance mode with benchtop NIR spectrometer (1000–2500 nm, 8 cm⁻¹ resolution, 64 scans).
• Preprocessing: Apply SNV followed by Savitzky-Golay first derivative (2nd order polynomial, 15-point window).
• Model Development: Build PLS-DA model using 70% of samples, with cross-validation.

Table 3: Performance Metrics for NIR Detection of Adulterants in Spices

Spice	Adulterant	Detection Level	Chemometric Method	Accuracy/Precision
Black Pepper	Starch, husks, spent pepper	5–10% (w/w)	PLS-DA	>90% classification accuracy [36]
Cumin	Allergenic nutshell powders	2–5% (w/w)	PCA-LDA	~95% sensitivity [36]
Saffron	Synthetic dyes, safflower	1–5% (w/w)	PLSR	R² > 0.95 [36]
Chili Powder	Sudan dyes	1–5 mg/kg	SIMCA	>90% specificity [36]
Ginger	Turmeric, starch	5–10% (w/w)	PLS-DA	>92% classification accuracy [36]

Nut Authentication

Nuts are vulnerable to adulteration with lower-quality varieties, foreign matter, or allergens. Cashew nuts, for example, may be adulterated with other nuts or fillers [38].

Experimental Protocol for Cashew Nut Authentication:

• Sample Preparation: Grunt nuts to consistent particle size (400 μm) using laboratory mill.
• Adulteration Design: Create binary, ternary, and quaternary mixtures with potential adulterants (e.g., peanuts, almonds, macadamia) at 5–30% concentration.
• Spectral Acquisition: Utilize both FT-NIR (1000–2500 nm) and ATR-FTIR (4000–400 cm⁻¹) for complementary data.
• Data Fusion: Implement mid-level data fusion of preprocessed features from both techniques.
• Model Development: Establish one-class SIMCA models for authentic cashew classification.

Data Interpretation: Effective detection of up to four adulterants in cashew nuts has been demonstrated using IR techniques combined with untargeted chemometrics and one-class SIMCA modeling, achieving high classification rates through data fusion approaches [38].

Liquid Food Authentication

Liquid foods like honey are frequently adulterated with sugar syrups, mislabeled regarding botanical origin, or diluted with water [30].

Experimental Protocol for Honey Authentication:

• Sample Preparation: Warm honey to 40°C to dissolve crystals, then equilibrate to 25°C. Homogenize thoroughly.
• Adulteration Design: Spike authentic honey with corn syrup, inverted sugar, or rice syrup at 5–25% (w/w) concentrations.
• Spectral Acquisition: Use transmission mode with 1 mm pathlength cell, 1100–2500 nm range, 32 scans.
• Preprocessing: Apply MSC followed by Savitzky-Golay smoothing (2nd order polynomial, 11-point window).
• Model Development: Develop PLSR models for quantitative prediction of adulteration levels.

Table 4: NIR Quantification of Honey Quality Parameters and Adulteration

Parameter	Spectral Range (nm)	Chemometric Method	Performance Metrics
Sugar Content (fructose/glucose)	1700–2100	PLSR	R² > 0.95 [30]
Moisture Content	1400–1500	PLSR	R² > 0.96 [30]
5-HMF	2200–2400	PLSR	R² > 0.90 [30]
Adulteration (syrups)	1000–2500	PCA-LDA	>90% classification accuracy [30]
Botanical Origin	Full spectrum	SIMCA	>85% correct classification [30]

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 5: Essential Research Reagents and Materials for NIR-Based Food Authentication

Item/Category	Specifications	Function/Application
Reference Materials	Certified authentic samples from verified sources	Establish baseline spectral libraries for model development
Sample Preparation	Laboratory mill (particle size <500 μm), moisture analyzer, standardized containers	Ensure sample uniformity and reproducibility
NIR Spectrometers	Benchtop (lab) and portable (field) devices; spectral range 1000–2500 nm	Spectral data acquisition under controlled and in-field conditions
Measurement Accessories	Reflectance cups, transmission cells (0.5–2 mm), fiber optic probes	Accommodate various sample types and physical forms
Chemometric Software	MATLAB with PLS_Toolbox, Unscrambler, R with chemometric packages	Data preprocessing, model development, and validation
Validation Standards	Independent sample sets with known adulteration levels	Model performance assessment and verification
3-Isocyanophenylformamide	3-Isocyanophenylformamide Research Chemical	3-Isocyanophenylformamide for research. A versatile isocyanide building block for multicomponent reactions. For Research Use Only. Not for human consumption.
D-Methionyl-L-serine	D-Methionyl-L-serine

Visualizing the Adulteration Detection Logic

The following diagram illustrates the logical decision process for classifying food samples as authentic or adulterated based on NIR spectral analysis:

This case study demonstrates that NIR spectroscopy, coupled with appropriate chemometric techniques, provides a powerful analytical framework for detecting adulteration in spices, nuts, and liquid foods. The methodology offers significant advantages over traditional techniques, including rapid analysis, minimal sample preparation, and comprehensive chemical profiling. As food supply chains grow increasingly complex, these non-destructive spectroscopic approaches will play an ever-more critical role in safeguarding food authenticity and protecting consumer interests. Future developments in portable instrumentation, machine learning integration, and expanded spectral libraries will further enhance the applicability of NIR spectroscopy for food authentication across diverse research and regulatory contexts.

In the realm of food quality and authenticity testing, infrared spectroscopy has emerged as a powerful analytical technique, offering rapid, non-destructive, and precise analysis of food constituents. This application note details protocols for using Near-Infrared Spectroscopy (NIRS) to predict functional and physicochemical properties in dairy and cereal matrices—two essential food categories with complex compositional profiles. The ability to rapidly assess parameters such as protein content, grain hardness, mycotoxin contamination, and dairy powder functionality addresses critical needs in quality control, product development, and safety assurance across the food industry [39] [40]. Within the broader context of thesis research on infrared spectroscopy for food quality, this document provides standardized methodologies that bridge conventional analytical techniques with modern chemometric innovations, enabling researchers to extract maximal information from spectral data while maintaining analytical rigor [39] [41].

Theoretical Foundations and Principles

Near-Infrared Spectroscopy operates in the electromagnetic radiation range of 780–2500 nm, corresponding to wavenumbers of approximately 12,500–4000 cm⁻¹ [39] [3]. This technique measures the absorption of infrared radiation by organic molecules, specifically focusing on the overtone and combination vibrations of hydrogen-containing functional groups such as C-H, O-H, N-H, and S-H [39] [40]. When NIR light interacts with a sample, these molecular bonds vibrate at frequencies characteristic of their chemical structure and environment, producing a unique spectral fingerprint that can be correlated with physicochemical properties through multivariate calibration [39] [42].

The fundamental equation governing this interaction is based on the energy of photons: E = h·f = h·c/λ where E represents energy, h is Planck's constant, f is frequency, c is the speed of light, and λ is wavelength [39]. The resulting spectra contain broad, overlapping absorption bands that require sophisticated chemometric tools for interpretation, making NIRS an indirect analytical method that depends on robust calibration models against reference analytical techniques [39] [40].

For cereal analysis, NIRS can assess diverse quality parameters including protein content, starch characteristics, grain hardness, and contamination markers [39] [40]. In dairy applications, it predicts compositional parameters (fat, protein, lactose) and functional properties such as solubility, bulk density, and ripening characteristics [40] [43]. The non-destructive nature of NIRS allows for repeated measurements of the same sample, enabling time-course studies of processes like cheese ripening without sample waste [43].

Experimental Protocols

Protocol 1: Cereal Quality Profiling via FT-NIRS

Scope: This protocol details the determination of protein content, hardness, and detection of mycotoxins (Aflatoxin B1) in whole wheat grains using Fourier-Transform Near-Infrared Spectroscopy (FT-NIRS).

• Sample Preparation:

  • Collect representative wheat samples (minimum 500 g) from different lots or varieties.
  • Clean samples to remove foreign materials and debris.
  • For whole grain analysis, ensure uniform moisture content (target ~12%) by conditioning if necessary.
  • For mycotoxin detection, artificially contaminate subsets with known concentrations of Aflatoxin B1 (AFB1) for calibration, validated with HPLC reference methods [44] [42].

• Spectral Acquisition:

  • Instrumentation: Use an FT-NIRS spectrometer equipped with a reflectance cup for whole grains or a transflectance cell for ground samples.
  • Parameters: Set scanning range to 1000–2500 nm, resolution to 8–16 cm⁻¹, and accumulate 64 scans per spectrum to optimize signal-to-noise ratio [39] [41].
  • Procedure: Fill the sample cup consistently, ensuring uniform packing. Acquire triplicate spectra from each sample, rotating the cup between scans. Include background scans every 30 minutes.

• Chemometric Analysis:

  • Preprocessing: Apply Standard Normal Variate (SNV) and Detrending to correct for scatter effects, followed by Savitzky-Golay first derivative (window: 11 points, polynomial order: 2) to enhance spectral features [39] [44].
  • Model Development: Use Partial Least Squares Regression (PLSR) for quantitative traits (protein, hardness). For mycotoxin classification, employ Principal Component Analysis (PCA) followed by Linear Discriminant Analysis (LDA) [39] [44].
  • Validation: Validate models using independent sample sets (not used in calibration) and report Root Mean Square Error of Prediction (RMSEP) and Coefficient of Determination (R²) [41].

Protocol 2: Dairy Powder Functionality Assessment

Scope: This protocol establishes a method for predicting solubility, bulk density, and free fat content in milk powder using a benchtop NIRS system.

• Sample Preparation:

  • Source milk powder samples with varied composition from different batches, brands, and storage conditions to ensure robust calibration [40].
  • Sieve powders to ensure consistent particle size (<250 µm) and store in airtight containers at 4°C until analysis to prevent moisture uptake.

• Spectral Acquisition:

  • Instrumentation: Use a benchtop NIR spectrometer with a powder analysis accessory.
  • Parameters: Configure the instrument to scan from 1100 to 2500 nm using a InGaAs detector. Collect spectra in reflectance mode.
  • Procedure: Present samples in a consistent, uniform depth in a sample cup. Compress gently with a standardized pressure plunger to ensure reproducible packing density. Acquire five spectra per sample, repacking between measurements.

• Chemometric Analysis:

  • Preprocessing: Apply Multiplicative Scatter Correction (MSC) to compensate for particle size effects, followed by vector normalization [40].
  • Model Development: Develop PLSR models correlating spectral data with reference values for each functional property (e.g., solubility index measured by the standard insolubility index method, bulk density by tapping method) [40].
  • Validation: Use cross-validation (e.g., leave-one-out or 10-fold) and external validation. Report the Ratio of Performance to Deviation (RPD), with RPD > 2.0 considered good for quantitative predictions [40].

Advanced Protocol: Transfer Learning for Mycotoxin Detection

Scope: This protocol utilizes transfer learning to adapt a pre-trained NIRS model for detecting Zearalenone (ZEN) in wheat, leveraging knowledge from an Aflatoxin B1 (AFB1) model to reduce the required target-domain sample size [44].

• Sample Preparation:

  • Prepare two sets: a large source set of wheat contaminated with AFB1 (n>200) and a smaller target set with ZEN (n=50).
  • Determine reference toxin concentrations using official LC-MS/MS methods [44].

• Spectral Acquisition:

  • Acquire spectra for both sets using the same NIR instrument and consistent settings as in Protocol 1.

• Model Development and Transfer:

  • Source Model: Train a deep learning model (e.g., 1D-Convolutional Neural Network) on the large AFB1 spectral dataset.
  • Transfer Learning: Freeze the initial layers of the pre-trained source model, which capture general spectral features. Replace and retrain the final layers using the smaller ZEN dataset to specialize the model for the new toxin [44].
  • Evaluation: Compare the performance (R², RMSEP) of the transfer-learned model against a model trained from scratch only on the ZEN data.

Data Presentation and Analysis

Table 1: Performance metrics of NIRS models for predicting key functional properties in cereals and dairy products.

Matrix	Property	NIR Region (nm)	Chemometric Method	R²	RMSEP	RPD	Reference
Wheat Grain	Protein Content	850-2500	PLSR	0.92-0.97	0.15-0.25%	2.5-4.1	[39] [41]
Wheat Grain	Hardness	1000-2500	PLSR	0.85-0.94	-	2.1-3.5	[40] [41]
Milk Powder	Solubility Index	1100-2500	PLSR	0.78-0.85	0.5-1.2%	1.8-2.3	[40]
Milk Powder	Bulk Density	1100-2500	PLSR	0.80-0.88	0.03-0.05 g/mL	2.0-2.5	[40]
Wheat	Aflatoxin B1 (AFB1)	1000-2500	Deep Learning (CNN)	>0.99	<0.5 ppb	>3.0	[44]
Cheese	Ripeness Stage	800-2500	PCA-LDA	0.90-0.95	-	-	[43]

Key Research Reagent Solutions

Table 2: Essential materials and reagents for NIRS-based quality profiling experiments.

Item	Specification / Function	Application Context
FT-NIRS Spectrometer	Wavelength range: 780-2500 nm; Detector: InGaAs; Integrating sphere or reflectance cup.	Primary instrument for spectral acquisition in cereal and powder analysis [39] [42].
Reference Materials	Certified standards for protein (e.g., Kjeldahl), mycotoxins (e.g., AFB1, ZEN).	Essential for developing and validating accurate calibration models [44] [41].
Quartz Sample Cups	Non-absorbing in NIR region; ensures consistent light path and packing.	Holding whole grains or powdered samples during scanning [39].
Chemometrics Software	Contains algorithms for PLSR, PCA, SVM, and pre-processing (SNV, MSC, Derivatives).	Critical for extracting meaningful information from complex spectral data [39] [3].
Portable NIR Spectrometer	Handheld device with fiber optic probe; suitable for in-field or at-line analysis.	Rapid screening of grain quality in storage or dairy powders in production [30] [3].

The protocols outlined in this application note demonstrate that NIRS is a robust and versatile technique for the quantitative prediction of functional and physicochemical properties in dairy and cereal products. Its speed, non-destructive nature, and minimal sample preparation offer significant advantages over traditional wet chemistry methods, enabling rapid quality control and supporting research and development.

Future developments in this field are likely to focus on several key areas. The integration of artificial intelligence and deep learning will further enhance model accuracy, especially for complex functional properties and trace-level contaminants [44] [3]. Transfer learning, as demonstrated in the advanced protocol, presents a powerful strategy to overcome the high cost of model development for new tasks, making NIRS applications more accessible and widespread [44]. Furthermore, the push for standardization and the creation of large, shared spectral libraries will improve model transferability between instruments and laboratories, fostering greater adoption of the technology in industrial settings [41] [43]. As these trends converge, NIRS will solidify its role as an indispensable tool for ensuring food quality, safety, and authenticity in the modern food industry.

Within the framework of infrared spectroscopy research for food quality and authenticity testing, hyperspectral imaging (HSI) and multispectral imaging (MSI) have emerged as transformative analytical techniques. These methods integrate the principles of spectroscopy and digital imaging, providing both spatial and spectral information in a single, non-destructive measurement [45] [46]. This capability is revolutionizing food analysis by enabling detailed characterization of chemical composition, physical structure, and external quality attributes simultaneously. While traditional infrared spectroscopy techniques, such as Near-Infrared (NIR) spectroscopy, excel at rapid quantitative analysis of major components like proteins and moisture, they lack spatial context [3] [47]. HSI and MSI bridge this gap, offering powerful tools for detecting contaminants, assessing quality, and verifying authenticity in complex food matrices, thereby supporting the core objectives of modern food research and industrial quality control.

Fundamental Principles and Technological Comparison

Core Principles of Hyperspectral and Multispectral Imaging

Hyperspectral and multispectral imaging are based on the interaction between light and matter across multiple wavelengths of the electromagnetic spectrum.

• Hyperspectral Imaging (HSI): HSI captures a contiguous spectrum for each pixel in an image, generating a three-dimensional data structure known as a hypercube. This hypercube comprises two spatial dimensions (x, y) and one spectral dimension (λ), where each spatial pixel contains a continuous, high-resolution spectrum [48] [46]. This allows for the detailed mapping of chemical constituents across a sample's surface.
• Multispectral Imaging (MSI): In contrast, MSI captures image data at a limited number of specific, discrete wavelengths [46]. While it provides less spectral detail than HSI, it is often sufficient for targeted applications and benefits from faster data acquisition and simpler processing.

The primary distinction lies in the number of wavelengths captured; MSI typically involves fewer than ten wavelengths, whereas HSI involves many more, often hundreds, providing a near-continuous spectral signature for each pixel [46].

Data Acquisition and Imaging Modes

The creation of a hypercube can be achieved through several scanning methodologies, each suited to different applications:

• Point Scanning (Whiskbroom): Captures the full spectrum of a single pixel at a time. The hypercube is built by scanning point-by-point across the sample. This method is highly precise but slower, making it suitable for static, high-precision laboratory analysis [48] [46].
• Line Scanning (Pushbroom): Captures a complete line of spatial pixels with their full spectra simultaneously. The hypercube is assembled as the sample or camera moves along the perpendicular direction. This mode is well-suited for the dynamic, on-line inspection of products on a conveyor belt [48] [46].
• Area Scanning (Wavelength Scanning): Captures a two-dimensional spatial image at one specific wavelength at a time. The hypercube is constructed by sequentially scanning across a range of wavelengths. This method is effective for applications requiring data from multiple specific wavelengths [48].

Data is typically acquired in one of three primary modes, depending on the relative positions of the light source, camera, and sample:

• Reflectance Mode: The light source and camera are on the same side of the sample. The camera captures light reflected from the sample's surface, which is widely used for assessing external quality attributes like color, surface defects, and contamination [48] [46].
• Transmittance Mode: The light source and camera are on opposite sides of the sample. The camera collects light that has passed through the sample, providing information about internal composition and defects [48].
• Interactance Mode: A combination of both, often using specific optics to separate the illuminating and reflected light paths on the same side of the sample [46].

Table 1: Key Characteristics of Hyperspectral and Multispectral Imaging

Feature	Hyperspectral Imaging (HSI)	Multispectral Imaging (MSI)
Spectral Resolution	High (Contiguous, narrow bands)	Low (Discrete, broad bands)
Number of Wavelengths	Many (Often hundreds)	Few (Typically <10)
Data Volume	Very large (Hypercube)	Moderate
Spectral Information	Full spectrum per pixel	Selective wavelengths per pixel
Primary Application	Research, complex quantification	Targeted, high-speed industrial inspection
Cost & Complexity	Higher	Lower

Applications in Food Quality and Authenticity Testing

The integration of spatial and spectral data enables a wide range of applications in food analysis, surpassing the capabilities of conventional spectroscopic methods.

Contaminant and Adulteration Detection

HSI is highly effective for identifying both biological and chemical contaminants. It has been successfully applied to detect fungal contamination, such as Penicillium digitatum in mandarins and Aspergillus niger in wheat, by identifying characteristic spectral changes associated with the infection [46]. Furthermore, HSI can identify non-conformities and adulteration, such as the detection of melamine in milk powder and the identification of adulterated oils, providing a rapid and non-destructive alternative to chromatographic methods [45] [3].

Quality and Compositional Analysis

The technology enables the quantitative prediction of key physicochemical properties in a wide range of food products. For instance, HSI can map the distribution and content of moisture, protein, fat, and carbohydrates in grains [48]. It can also assess quality parameters in meat, such as color, pH, tenderness, and drip loss [46]. For fruits and vegetables, HSI can monitor internal attributes like soluble solids content and even detect anthocyanin levels in grapes [46].

Geographic Origin and Authenticity Tracing

By analyzing the unique spectral fingerprints influenced by growing conditions, HSI and NIR spectroscopy can be used for geographic origin verification. Studies have demonstrated the feasibility of tracing the origin of products like tea oil and milk by combining spectral data with machine learning classifiers such as Support Vector Machines (SVM) and Convolutional Neural Networks (CNN), achieving high prediction accuracies [3] [48].

Table 2: Representative Applications of HSI and MSI in Food Analysis

Application Area	Specific Example	Key Findings/Performance
Disease & Contamination	Detection of fungal contamination in wheat [48]	Identification based on chlorophyll degradation at 680 nm.
Adulteration Detection	Identification of melamine in milk powder [46]	Non-destructive detection and quantification of adulterant.
Physicochemical Analysis	Prediction of protein & moisture in grains [48]	Established regression models between spectral data and chemical parameters.
Meat Quality	Assessment of beef tenderness and color [46]	Correlation between spectral features and quality parameters.
Geographical Origin	Tracing origin of milk [3]	Portable NIR with FDLDA-KNN classifier achieved 97.33% accuracy.
Mycotoxin Detection	Aflatoxin B1 in maize [46]	Potential for non-destructive screening of mycotoxins.

Experimental Protocols

This section provides a detailed methodology for a typical HSI-based experiment aimed at assessing food quality and authenticity, using grain quality assessment as a model application.

Protocol 1: Hyperspectral Imaging for Grain Quality Assessment

1. Objective To non-destructively predict the protein content and detect fungal contamination in wheat kernels using a line-scanning HSI system in the visible and near-infrared (VNIR) range.

2. Materials and Reagents

• Grain samples: Whole wheat kernels (~200 g).
• Reference analytical equipment: Kjeldahl apparatus or Dumas combustion analyzer for protein content validation.
• Sample presentation: Black, non-reflective tray or conveyor belt.

3. Hyperspectral Image Acquisition

• Instrument Setup: Use a line-scanning HSI system equipped with a spectrograph covering the 400-1000 nm range (e.g., ImSpector V10E) and a CCD camera [48].
• Acquisition Parameters:
  • Illumination: Halogen lamps (e.g., 2×150 W) positioned at a 45° angle to the sample.
  • Camera lens-to-sample distance: Calibrated to achieve desired spatial resolution.
  • Exposure time: Optimized to avoid pixel saturation (e.g., 10-30 ms).
  • Conveyor speed: Synchronized with camera frame rate for sharp images.
• Scanning: Place kernels in a single layer on the conveyor. Acquire hyperspectral images of the samples in reflectance mode. Capture calibration images: a white reference (≥99% reflectance standard) and a dark reference (with lens covered) for radiometric correction [48] [46].

4. Data Preprocessing

• Radiometric Correction: Convert raw digital numbers to relative reflectance (R) using the formula: ( R = (I{sample} - I{dark}) / (I{white} - I{dark}) ) where ( I ) is the intensity.
• Spectral Preprocessing: Apply algorithms to minimize noise and scattering effects:
  • Savitzky-Golay (SG) Filtering: For smoothing and noise reduction [48].
  • Standard Normal Variate (SNV): To correct for scattering effects caused by uneven surfaces [3] [48].
  • Derivative Methods (e.g., 1st or 2nd derivative): To enhance spectral features and resolve overlapping peaks [3] [48].

5. Feature Wavelength Selection

• To reduce data dimensionality and model complexity, identify the most informative wavelengths related to protein and fungal contamination.
• Methods:
  • Principal Component Analysis (PCA): For unsupervised exploration and identification of wavelengths with the highest variance [45] [48].
  • Competitive Adaptive Reweighted Sampling (CARS): A method that selects wavelengths with strong correlations to the component of interest [3].

6. Model Development and Validation

• Reference Data: Determine the actual protein content of each kernel using the standard Kjeldahl method. For fungal contamination, use microbiological plating as a reference.
• Quantitative Model (for protein):
  • Use Partial Least Squares Regression (PLSR) to establish a model between the preprocessed spectral data and the reference protein values [45] [48].
  • Randomly split samples into a calibration set (e.g., 70%) and a validation set (e.g., 30%).
• Qualitative Model (for contamination):
  • Use Support Vector Machine (SVM) or Linear Discriminant Analysis (LDA) to classify kernels as "healthy" or "infected" [45] [3].
• Model Evaluation: Assess the PLSR model using the Coefficient of Determination (R²) and the Root Mean Square Error (RMSE) of calibration and validation. Assess the classification model using overall accuracy.

Protocol 2: NIR Spectroscopy for Liquid Food Adulteration

1. Objective To rapidly detect and quantify the level of adulteration in edible oil using a Fourier-Transform Near-Infrared (FT-NIR) spectrometer.

2. Materials and Reagents

• Samples: Pure peanut oil and potential adulterants (e.g., lower-cost oils).
• Preparation: Create adulterated samples with known concentration gradients (0-30% v/v).
• Cuvettes: Quartz cuvettes suitable for NIR transmission measurements.

3. Spectral Collection

• Instrument: Use an FT-NIR spectrometer equipped with a transmission cell.
• Acquisition:
  • Collect spectra over the wavenumber range of 12,820–4,000 cm⁻¹ (approx. 780–2500 nm) [3].
  • For each sample, take multiple scans (e.g., 32) and average them to improve the signal-to-noise ratio.
  • Maintain a constant temperature during measurement.

4. Data Preprocessing and Modeling

• Preprocessing: Apply S-G smoothing and first-order derivatives to enhance spectral features [3].
• Modeling: Develop a Partial Least Squares (PLS) regression model to correlate the spectral data with the adulteration concentration.
• Validation: Use cross-validation (e.g., leave-one-out) to evaluate model performance, reported as R² and RMSECV (Root Mean Square Error of Cross-Validation).

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Key Research Reagents and Equipment for HSI/NIR Experiments

Item	Function/Description	Example Specifications
Hyperspectral Imaging System	Core instrument for capturing spatial and spectral data.	Includes camera, spectrograph (e.g., 400-1000 nm), and optics [48].
NIR Spectrometer	For rapid collection of spectral data without spatial resolution.	Portable or benchtop FT-NIR spectrometer [3].
Halogen Lamp	Broadband illumination source for VNIR HSI systems.	100-300W tungsten halogen lamp, 340-2500 nm range [46].
Calibration Standards	Essential for radiometric and wavelength calibration.	White reference (e.g., Spectralon), dark reference, and wavelength standard [46].
Chemometrics Software	For data preprocessing, modeling, and visualization.	MATLAB, Python (with scikit-learn, NumPy), or commercial software (e.g., Unscrambler) [48].
Reference Analytical Equipment	To obtain ground truth data for model calibration.	HPLC for mycotoxins, Kjeldahl for protein, GC for fatty acids [3] [46].
2-(2-Methylpropyl)azulene	2-(2-Methylpropyl)azulene	2-(2-Methylpropyl)azulene is a high-purity azulene derivative for research applications. Explore its potential in anti-inflammatory and material science studies. For Research Use Only. Not for human or veterinary use.
Kanzonol H	Kanzonol H CAS 152511-46-1 - Licorice Flavonoid	High-purity Kanzonol H, a prenylated flavonoid from licorice. Explore its research applications in wound healing and inflammation. For Research Use Only. Not for human consumption.

Data Processing and Analysis Workflow

The analysis of HSI data is a multi-stage process that relies heavily on chemometrics. A standard workflow is illustrated in Figure 1 and detailed below.

1. Preprocessing: Raw hyperspectral data contains noise and artifacts from the instrument and environment. Preprocessing aims to enhance the meaningful signal. Key techniques include:

• Noise Reduction: Savitzky-Golay filtering smooths the spectra [48].
• Scatter Correction: Multiplicative Scatter Correction (MSC) and Standard Normal Variate (SNV) correct for light scattering due to uneven surfaces [3] [48].
• Spectral Derivatization: First- and second-order derivatives are used to remove baseline offsets and resolve overlapping peaks [48].

2. Feature Extraction: A full HSI cube contains a massive amount of data, much of which is redundant. Feature extraction identifies the most informative wavelengths related to the property of interest, significantly reducing data dimensionality. Common methods include Principal Component Analysis (PCA) and more advanced techniques like Competitive Adaptive Reweighted Sampling (CARS) [3] [48].

3. Model Development: Mathematical models are built to correlate spectral features with reference measurements.

• Quantitative Analysis: Partial Least Squares Regression (PLSR) is the most widely used method for predicting continuous variables, such as protein content or adulteration levels [45] [3].
• Qualitative Analysis: For classification tasks (e.g., infected vs. healthy, authentic vs. adulterated), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Soft Independent Modelling of Class Analogy (SIMCA) are frequently employed [45] [3].

The integration of deep learning, particularly Convolutional Neural Networks (CNNs), is an emerging trend that automates feature extraction and modeling, often leading to improved accuracy in tasks like geographical origin tracing [3] [47].

Optimizing Your Analysis: Overcoming Challenges in IR Spectroscopy

Infrared (IR) spectroscopy has emerged as a powerful analytical tool for ensuring food quality and authenticity, particularly suited for analyzing complex matrices like spices and powdered foods. These samples present significant analytical challenges due to their inherent heterogeneity, variable particle sizes, and susceptibility to environmental factors such as moisture. The physical structure of powdered foods facilitates fraudulent practices, making them vulnerable to adulteration with lower-cost materials such as starches, flours, rice by-products, and even allergenic nutshells [33] [49]. Such adulteration not only causes economic losses but also poses public health risks, including allergic reactions and exposure to toxic substances [33]. Navigating this complexity requires robust, non-destructive analytical strategies that can handle sample variability while providing accurate authentication. IR spectroscopy, coupled with advanced chemometrics, offers rapid, non-destructive analysis essential for modern food quality control frameworks, enabling detection of fraud, verification of origin, and identification of potential contaminants [33] [50].

Fundamental Challenges in Analyzing Complex Matrices

The analysis of spices, powders, and other heterogeneous food matrices via IR spectroscopy is complicated by several physical and chemical factors that can significantly impact spectral quality and analytical results. Particle size distribution stands as a primary challenge, as uneven particles cause inconsistent light scattering, leading to baseline shifts and multiplicative scattering effects that obscure meaningful chemical information [33]. The moisture content in hygroscopic powders affects the intensity of O-H stretching bands, potentially masking signals from other constituents [33]. Furthermore, heterogeneous composition creates sampling representativeness issues, where a single spectrum may not accurately reflect the overall sample composition [33]. Environmental factors such as probe-to-sample distance, measurement angle, and packaging materials introduce additional spectral variations unrelated to chemical composition [33]. These matrix-specific effects manifest as baseline shifts, slope changes, and heightened spectral noise, necessitating comprehensive sample preparation strategies and advanced spectral preprocessing to extract reliable chemical information [33].

Experimental Protocols for Sample Preparation and Analysis

Sample Preparation Standardization

Proper sample preparation is crucial for obtaining reproducible IR spectra from complex matrices. The following standardized protocol ensures minimal spectral variance due to physical sample characteristics:

• Moisture Control: Condition all samples in a controlled environment (relative humidity: 40-50%, temperature: 20-25°C) for 24 hours before analysis to standardize water activity [33]. For hygroscopic materials, use desiccators with standardized drying agents.

• Particle Size Standardization: Grind solid samples using laboratory mills equipped with standardized sieve systems. For most powdered foods and spices, a particle size range of 150-250 µm provides optimal spectral reproducibility [33]. Verify particle size distribution using laser diffraction methods.

• Homogenization Procedure: Employ geometric mixing techniques (coning and quartering) for at least 5 minutes to ensure uniform composition. For laboratory samples, use a V-blender for 10-15 minutes for thorough homogenization [49].

• Packaging and Storage: Store prepared samples in airtight, light-resistant containers at constant temperature (4°C for long-term storage) to prevent compositional changes. Allow samples to reach room temperature before analysis [33].

Spectral Acquisition Parameters

Optimized instrument parameters ensure consistent spectral collection across different sample types:

Table 1: Recommended Spectral Acquisition Parameters for Different IR Techniques

Parameter	FTIR-ATR	NIR Spectroscopy (Diffuse Reflectance)	NIR Hyperspectral Imaging
Spectral Range	4000-400 cm⁻¹	10000-4000 cm⁻¹ (1000-2500 nm)	900-1700 nm (portable)
Resolution	4 cm⁻¹	8-16 cm⁻¹	5-10 nm
Scan Number	32-64 scans	16-32 scans	Varies by spatial resolution
Apodization	Happ-Genzel	Happ-Genzel	Not applicable
Sample Contact Pressure	Consistent firm pressure	Not applicable	Not applicable
Sample Presentation	Direct contact with ATR crystal	Quartz sample cups	Conveyor belt or stage

For FTIR-ATR analysis, ensure consistent pressure application between the sample and crystal using the instrument's pressure applicator [51]. For NIR analysis of powders, maintain consistent sample cup filling volume and tamping pressure [33]. For heterogeneous samples, collect multiple spectra from different sample positions and average them to improve representativeness [49].

Chemometric Data Processing Workflows

The analysis of complex matrices requires sophisticated chemometric approaches to extract meaningful information from IR spectra. The following workflow outlines the standard procedure for model development:

Spectral Preprocessing Strategies

Preprocessing corrects for physical artifacts and enhances chemical information:

Table 2: Spectral Preprocessing Techniques for Complex Matrices

Technique	Primary Function	Application Context	Effect on Spectra
Savitzky-Golay Smoothing	Reduces high-frequency noise	All powder and spice matrices	Improves signal-to-noise ratio
Standard Normal Variate	Corrects scattering effects	Heterogeneous particle distributions	Removes multiplicative interferences
Multiplicative Scatter Correction	Compensates for scattering	Powders with varying densities	Corrects additive and multiplicative effects
First Derivative	Removes baseline shifts	Overlapping absorption bands	Emphasizes subtle spectral features
Second Derivative	Enhances band resolution	Complex multicomponent mixtures	Resolves overlapping peaks
Detrending	Eliminates curvilinear baselines	Samples with varying particle size	Removes wavelength-dependent scattering

The selection and sequence of preprocessing techniques significantly impact model performance. A common effective combination for powdered foods includes Savitzky-Golay smoothing (window: 11 points, polynomial order: 2) followed by Standard Normal Variate transformation [33] [49].

Classification and Regression Models

Following preprocessing, various chemometric models enable qualitative and quantitative analysis:

• Principal Component Analysis: An unsupervised method for exploring natural clustering and detecting outliers. Essential for initial data exploration to identify patterns and potential anomalies in complex datasets [33] [50].

• Partial Least Squares Regression: The most widely used regression method for quantitative analysis in IR spectroscopy. Particularly effective for predicting adulteration levels in powdered foods, with reported R² values >0.97 for cumin adulterated with rice by-products [49].

• Support Vector Machines: Powerful for non-linear classification problems, such as authenticating geographical origin or detecting specific adulterants in complex matrices [33] [50].

• Artificial Neural Networks: Including Multi-Layer Perceptron and Long Short-Term Memory networks, these deep learning approaches show superior performance for interpreting complex spectral data, with RMSE values <1.3% reported for adulteration quantification [49].

The model development process follows a logical progression from data acquisition through validation, as illustrated below:

Chemometric Analysis Workflow for Complex Matrices

Essential Research Reagent Solutions

Successful implementation of IR spectroscopy for analyzing complex matrices requires specific materials and reagents to ensure analytical rigor:

Table 3: Essential Research Materials for IR Spectroscopy of Complex Matrices

Material/Reagent	Function	Application Example
Certified Reference Materials	Method validation and calibration	Authentic spice samples for model development
Silica Gel Desiccant	Moisture control in hygroscopic samples	Standardizing powder moisture content before analysis
Standardized Sieve Sets	Particle size control	Achieving uniform particle distribution (150-250 µm)
ATR Cleaning Solvents	Crystal maintenance	High-purity ethanol and acetone for FTIR-ATR
Background Reference Materials	Instrument calibration	Spectralon for diffuse reflectance, empty chamber for transmission
Adulterant Standards	Model training	Rice bran, cassava starch, nutshell powders for adulteration studies

Application Case Study: Detection of Rice By-Products in Cumin Powder

A comprehensive case study demonstrates the practical application of these strategies for detecting adulteration in cumin powder with rice by-products (rice bran and small broken rice) [49]:

Experimental Design

• Sample Preparation: Pure cumin samples were adulterated with rice bran and small broken rice at concentrations ranging from 5% to 50% (w/w). Samples were homogenized using a laboratory blender for 15 minutes and sieved through a 250 µm mesh.

• Spectral Acquisition: NIR spectra were collected in diffuse reflectance mode across 1000-2500 nm range with 8 cm⁻¹ resolution. Sixty-four scans were averaged for each spectrum, with three replicates per sample.

• Data Analysis: Multiple preprocessing techniques were applied including SNV, Savitzky-Golay smoothing, and first derivative. Models were developed using PLSR, MLP, and LSTM approaches with k-fold cross-validation.

Results and Performance Metrics

The analysis demonstrated excellent predictive capability for quantifying adulteration levels:

Table 4: Performance Comparison of Chemometric Models for Cumin Adulteration Detection

Model Type	Adulterant	R²	RMSE (%)	Key Advantages
PLSR	Rice Bran	0.981	3.12	Interpretability, computational efficiency
PLSR	Small Broken Rice	0.974	3.85	Stability with limited samples
MLP	Rice Bran	0.994	1.28	Superior non-linear modeling
MLP	Small Broken Rice	0.991	1.52	High predictive accuracy
LSTM	Rice Bran	0.985	2.74	Temporal pattern recognition
LSTM	Small Broken Rice	0.982	3.01	Sequential data processing

Notably, PCA enabled clear separation of pure and adulterated samples at levels as low as 5%, demonstrating the sensitivity of these approaches for detecting economically-motivated adulteration [49].

Emerging Trends and Future Perspectives

The field of IR spectroscopy for complex matrix analysis continues to evolve with several promising developments:

• Portable and Miniaturized Devices: Compact NIR and FTIR instruments enable on-site analysis at multiple points in the supply chain, facilitating real-time authentication decisions [33] [37]. Recent advances have improved the performance of these devices, making them viable alternatives to benchtop systems for many applications.

• Advanced Deep Learning Architectures: Convolutional Neural Networks and sophisticated neural networks are increasingly applied to spectral data, demonstrating superior performance for complex pattern recognition tasks compared to traditional chemometrics [49] [50].

• Data Fusion Approaches: Combining multiple spectroscopic techniques (e.g., NIR with FTIR) or integrating spectral data with other analytical measurements provides complementary information that enhances authentication capability [50].

• Self-Adaptive Chemometric Models: Development of algorithms that can continuously learn and adapt to new sample variations, reducing the need for frequent model recalibration [33].

These advancements position IR spectroscopy as an increasingly powerful tool for authenticating complex food matrices, with potential applications expanding to real-time monitoring throughout production processes and supply chains.

In the field of infrared (IR) spectroscopy for food quality and authenticity testing, the recorded raw spectra are not immediately suitable for analysis. They are often laden with various non-chemical spectral distortions that can obscure vital chemical information and compromise the accuracy of subsequent chemometric models [52]. Data preprocessing is therefore a critical first step in the chemometric workflow, serving to remove these unwanted variations and enhance the genuine molecular features of the sample [52] [53]. For research aimed at distinguishing authentic food products from adulterated ones or quantifying key quality parameters, proper preprocessing is indispensable for building robust and reliable predictive models [2] [54]. This document outlines the core techniques and protocols for effective data preprocessing, specifically within the context of food analysis.

Core Preprocessing Techniques

The primary goals of preprocessing are to mitigate physical and instrumental artifacts, including light scattering, baseline shifts, and random noise. The following techniques are fundamental to achieving these goals.

Scatter Correction

Scattering effects, often caused by variations in particle size or surface roughness in solid food samples, manifest as multiplicative and additive effects in spectra, overshadowing the desired chemical absorbance data [2] [52].

• Multiplicative Scatter Correction (MSC): This is a model-based method that corrects for both additive and multiplicative scattering effects. Each spectrum is corrected by using the average spectrum of the dataset as a reference [2].
• Standard Normal Variate (SNV): This method operates on each individual spectrum, centering the data (correcting baseline shifts) and then scaling it to correct for the variations in optical path length [2] [52]. SNV is particularly useful when an average reference spectrum is not appropriate.

Table 1: Comparison of Primary Scatter Correction Techniques

Technique	Principle	Primary Use Case	Key Advantage
Multiplicative Scatter Correction (MSC)	Corrects each spectrum based on a reference (often the mean spectrum) to remove additive and multiplicative effects [2].	Solid samples with varying particle sizes (e.g., powdered milk, ground coffee) [2].	Effective for datasets where all samples have a similar chemical composition and scattering is the main variation.
Standard Normal Variate (SNV)	Centers and scales each spectrum independently, line by line [2] [52].	Similar to MSC, but more suitable when a stable reference spectrum is not available.	Treats each spectrum individually, making it robust for more heterogeneous sample sets.

Baseline Alignment and Correction

Baseline distortions—offsets, slopes, or curvature—can arise from instrumental drift, light scattering, or sample matrix effects [52] [53]. Correction is essential to ensure that absorbance values accurately reflect chemical concentration.

• Derivative Techniques: Applying first or second derivatives (e.g., using the Savitzky-Golay algorithm) is a highly effective method for resolving overlapping peaks and simultaneously removing baseline offsets and linear slopes [2]. A key consideration is that derivative processing can amplify high-frequency noise, which is often first mitigated by smoothing [2].
• Straight Line Subtraction (SLS): This algorithm fits a straight line to the spectrum and subtracts these values from the original spectrum to correct for baseline deviations [2].

Noise Reduction

Reducing random noise is crucial for improving the signal-to-noise ratio and the stability of chemometric models.

• Smoothing: The Savitzky-Golay filter is the most widely used smoothing algorithm. It fits a low-degree polynomial to successive windows of data points, effectively smoothing the data without significantly distorting the signal shape, including peak heights and widths [2] [55].
• Normalization: This step adjusts all spectra to a common intensity scale, compensating for differences in sample quantity or pathlength. Common methods include dividing by the most intense peak (peak normalization) or by the total absorbance area (area normalization) [52] [53].

Integrated Experimental Protocols

Protocol 1: Standard Workflow for FT-IR ATR Analysis of Liquid Foods

This protocol is designed for analyzing liquid food samples like milk, juice, or oil using a Fourier-Transform Infrared spectrometer with an Attenuated Total Reflection (ATR) accessory [52] [54].

1. Sample Presentation:

• Ensure the ATR crystal (e.g., diamond) is clean. Gently wipe with a soft cloth moistened with ethanol and allow to dry completely.
• Apply a representative aliquot of the liquid sample directly onto the crystal, ensuring the crystal is fully covered.
• Lower the pressure clamp to ensure uniform and optimal contact between the sample and the crystal. For consistent results, use a fixed torque setting if available [54].

2. Spectral Acquisition:

• Collect spectra over the appropriate wavenumber range (e.g., 4000–800 cm⁻¹ for MIR).
• Set the scanner velocity, resolution (typically 4–8 cm⁻¹), and co-add a sufficient number of scans (e.g., 32–64) to achieve an adequate signal-to-noise ratio [54].
• Collect a background spectrum (without sample) under identical instrument settings before measuring each sample or batch.

3. Data Preprocessing Sequence: The following sequence is recommended as a starting point for liquid food authentication [52] [3]:

• Mean Centering: Subtract the average value of each variable (wavenumber) across all spectra. This centers the data and facilitates clearer model interpretation in Principal Component Analysis (PCA) [52].
• Scatter Correction: Apply SNV to correct for pathlength differences and scattering effects [52].
• Smoothing: Apply a Savitzky-Golay filter (e.g., 2nd-order polynomial, 11–15 points) to reduce high-frequency noise [55].
• Derivatization: Apply a first or second derivative (e.g., using the Savitzky-Golay algorithm) to resolve overlapping peaks and remove residual baseline shifts [2] [52].

Protocol 2: Developing a Calibration Model for Quantitative Analysis

This protocol details the steps for creating a calibration model to predict a specific constituent, such as protein content in meat or geographic origin of honey [2] [55].

1. Experimental Design and Reference Analysis:

• Assemble a representative set of calibration samples that cover the full expected range of the property of interest.
• Analyze these samples using a primary reference method (e.g., Kjeldahl method for protein) to obtain reference values [2]. The accuracy of the final NIR model is dependent on the accuracy of this reference data.

2. Spectral Preprocessing and Model Development:

• Acquire spectra for all calibration samples.
• Test different preprocessing combinations (e.g., SNV+Detrend, 1st Derivative, 2nd Derivative) to determine the optimal strategy for your specific dataset [52].
• Use Partial Least Squares Regression (PLSR) to build a model that correlates the preprocessed spectral data (X-matrix) with the reference analytical data (Y-matrix) [2] [56].
• Validate the model using an independent set of validation samples not used in the calibration. Evaluate performance using metrics like the coefficient of determination (R²), Root Mean Square Error of Prediction (RMSEP), and Mean Absolute Error [55].

Workflow Visualization

The following diagram illustrates the logical workflow for preprocessing infrared spectral data, from raw acquisition to a model-ready dataset.

The Scientist's Toolkit: Key Reagent and Material Solutions

Table 2: Essential Materials for Infrared Spectroscopy in Food Analysis

Item	Function/Application
ATR Crystals (Diamond, ZnSe)	The internal reflection element in ATR accessories. Diamond is durable and chemically inert, ideal for heterogeneous samples. ZnSe offers a good balance of performance and cost but is susceptible to acidic damage [54].
Solvents for Cleaning (Ethanol, HPLC-grade water)	Used to clean the ATR crystal between samples to prevent cross-contamination. Must be volatile to leave no residue [54].
Background Standards	Materials used for collecting a background spectrum. For ATR, this is typically the clean, empty crystal. For transmission, a blank cell filled with solvent is used.
Certified Reference Materials	Food matrices with certified compositional data. Essential for validating the accuracy of spectroscopic methods and calibration models [2].
Savitzky-Golay Filter	A digital filter that can be applied for both smoothing and calculating derivatives of spectral data, fundamental for noise reduction and baseline correction [2] [55].
Iridium--oxopalladium (1/1)	Iridium--oxopalladium (1/1)|CAS 142261-85-6|RUO

In the field of infrared (IR) spectroscopy for food quality and authenticity testing, the transition from a robust laboratory model to a reliable deployed method presents two significant challenges: avoiding model overfitting and ensuring successful calibration transfer between instruments. Near-infrared (NIR) spectroscopy, combined with machine learning, has emerged as a powerful technique for rapid, non-destructive analysis of food products, enabling real-time quality assessments with minimal sample preparation [57]. The application of this technique spans various domains, from determining flavonoid and protein content in buckwheat to screening liquid foods for adulteration and verifying their geographic origin [57] [3]. However, the predictive performance of these models depends critically on their robustness—their ability to maintain accuracy when faced with new samples, different environmental conditions, or alternative instrumentation. This application note provides detailed protocols and data-driven guidance for developing robust spectroscopic models and effectively transferring them across devices, with a specific focus on food authenticity applications.

Theoretical Background

The Overfitting Problem in Spectroscopic Modeling

Overfitting occurs when a model learns not only the underlying relationship between spectral features and analyte concentration but also the noise and random variations present in the training dataset. Such a model typically exhibits excellent performance on the training data but fails to generalize to new, unseen samples. This is particularly problematic in spectroscopy, where datasets often contain a large number of spectral variables (wavelengths) relative to a limited number of sample observations, creating a high-dimensional modeling environment prone to this issue.

The Calibration Transfer Challenge

Calibration transfer addresses the problem of maintaining model performance when a calibration developed on a primary (master) instrument is applied to spectral data collected from a secondary (slave) instrument. Even instruments of the same model and manufacturer exhibit subtle differences in optical components, detectors, or environmental conditions, leading to spectral variations that can severely degrade model performance if not corrected [58]. Effective calibration transfer is essential for scalable deployment of spectroscopic methods across multiple instruments at different locations in the food supply chain.

Experimental Protocols

Protocol for Developing Robust, Non-Overfit Models

Principle: Implement a rigorous validation framework and strategic data preprocessing to ensure models capture genuine chemical information rather than instrumental noise.

Materials:

• Training set spectra with reference values
• Independent validation set not used in model training
• Chemometric software (e.g., SIMCA, MATLAB, Python with scikit-learn)
• Standard Normal Variate (SNV) and Savitzky-Golay filtering algorithms

Procedure:

• Sample Preparation and Spectral Acquisition:
  • Collect a representative set of samples covering expected biological and compositional variability. For buckwheat analysis, this included 60 seed samples (30 Tartary and 30 common) sourced from a single supplier [57].
  • Perform meticulous sample preparation including cleaning, grinding, drying, and consistent storage to minimize non-chemical spectral variance [57].
  • Acquire spectra using a calibrated NIR spectrometer (e.g., NIR1700 spectrometer, 900-1700 nm, 32 scans per measurement) [57].

• Data Preprocessing:

  • Apply appropriate spectral preprocessing techniques to remove physical light scattering effects and enhance chemical signals.
  • Test multiple preprocessing methods including SNV, Multiplicative Scatter Correction (MSC), and Savitzky-Golay derivatives [57] [3].
  • For NIR data, employ feature selection techniques like Successive Projections Algorithm (SPA) to identify the most informative wavelengths and reduce dimensionality [57].

• Model Training with Embedded Validation:

  • Split data into training (≈70%), validation (≈15%), and test sets (≈15%) prior to model development.
  • Test multiple algorithm types: Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Backpropagation Neural Networks (BPNN) [57].
  • Implement hyperparameter optimization using techniques like Particle Swarm Optimization (PSO) for SVR models [57].
  • For each model configuration, monitor performance on the validation set during training to detect overfitting.

• Model Evaluation and Selection:

  • Evaluate final models on the completely held-out test set.
  • Select the model with the best performance on both validation and test sets, not the training set.
  • For flavonoid prediction in buckwheat, the RAW-SPA-CV-SVR model achieved optimal performance (R²p = 0.9811, RMSEP = 0.1071) [57].

Table 1: Performance Metrics for Robust Model Selection in Buckwheat Analysis

Model Type	Application	Preprocessing	R²p	RMSEP
RAW-SPA-CV-SVR	Flavonoid prediction	RAW, SPA	0.9811	0.1071
MMN-SPA-PSO-SVR	Protein prediction	MMN, SPA, PSO	0.9247	0.3906
PLSR	Flavonoid prediction	Multiple tested	Lower than SVR	Higher than SVR
BPNN	Flavonoid prediction	Multiple tested	Lower than SVR	Higher than SVR

Protocol for Calibration Transfer Between Instruments

Principle: Use spectral standardization algorithms to correct for instrumental differences, enabling a single model to be applied across multiple devices.

Materials:

• Primary (master) spectrometer
• Secondary (slave) spectrometer(s) of same model
• Standard reference materials (e.g., Spectralon)
• Set of transfer samples (20-30) covering spectral space of interest
• Software capable of spectral standardization (e.g., with DS, POS algorithms)

Procedure:

• Instrument Synchronization:
  • Establish identical instrument parameters on both primary and secondary devices (scan time, optical gain, spectral range) [58].
  • Conduct warm-up procedures (e.g., 10 minutes continuous operation) before data collection to stabilize measurements [58].
  • Collect background measurements regularly using a consistent reference standard (e.g., Spectralon with 99% reflectance) [58].

• Transfer Sample Selection and Measurement:

  • Select 20-30 transfer samples representing the full spectral variability of the application domain.
  • Measure these samples on both the primary and secondary instruments in randomized order to avoid systematic bias.
  • For oregano authentication, researchers utilized 295 authentic oregano samples and 109 potential adulterant samples to develop and validate transferable models [58].

• Standardization Model Development:

  • Test multiple standardization approaches including Direct Standardisation (DS), Piecewise Direct Standardisation (PDS), and Orthogonal Signal Correction (OSC) [58].
  • Develop transformation matrices that map secondary instrument spectra to the primary instrument's spectral space.
  • Evaluate different standardization performance by comparing model predictions on the secondary device before and after correction.

• Model Transfer and Validation:

  • Apply the standardization model to correct spectra collected on the secondary instrument.
  • Use the original classification or quantification model (developed on the primary instrument) to analyze corrected spectra from the secondary instrument.
  • Validate performance with an independent set of samples not used in standardization development.
  • For oregano authentication, the optimized model correctly predicted 90% of authentic oregano and 100% of adulterant samples on a secondary device without recalibration [58].

Table 2: Calibration Transfer Performance for Oregano Authentication

Standardization Method	Primary Device Correct Prediction	Secondary Device Correct Prediction
Raw (no standardization)	93.0% (Oregano), 97.5% (Adulterants)	90.0% (Oregano), 100% (Adulterants)
Direct Standardisation (DS)	Not Reported	Performance Varies
Piecewise Direct Standardisation (PDS)	Not Reported	Performance Varies
Orthogonal Signal Correction (OSC)	Not Reported	Performance Varies

Data Visualization and Workflows

Model Development Workflow

Calibration Transfer Workflow

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential Materials for Robust Spectroscopic Analysis

Item	Specification/Example	Function/Application
Portable NIR Spectrometer	NeoSpectra Micro, SciAps vis-NIR, Metrohm TaticID-1064ST [59] [58]	Field-deployable analysis for supply chain screening
FT-IR Spectrometer	Bruker Vertex NEO with vacuum ATR [59]	High-precision laboratory analysis, removes atmospheric interference
Spectral Reference Standard	Spectralon (99% reflectance) [58]	Regular instrument calibration and background measurement
Chemometrics Software	SIMCA, Python/R with scikit-learn/caret [58]	Data preprocessing, model development, and validation
Sample Preparation Equipment	Grinders, drying ovens, controlled storage containers [57]	Ensure sample consistency and minimize physical spectral variance
Reference Analytical Equipment	GC-MS, HPLC [3]	Provide reference values for model training and validation

Developing robust spectroscopic models that avoid overfitting and successfully transfer across instruments requires meticulous attention to experimental design, data preprocessing, and validation strategies. The protocols outlined herein provide a framework for creating models that maintain predictive accuracy in real-world food authenticity applications. As spectroscopic technology continues to evolve toward portable, field-deployable devices [59] [58], these robustness principles become increasingly critical for ensuring reliable food quality monitoring throughout complex global supply chains. Future work should focus on enhancing model interpretability, developing more efficient transfer learning approaches, and establishing standardized protocols for specific food commodity applications.

In the field of infrared spectroscopy for food quality and authenticity testing, the analytical success of any method hinges on two critical decisions: selecting the appropriate wavelength region and choosing the optimal algorithm for data processing. Infrared spectroscopy provides a rapid, non-destructive means of assessing the chemical composition of food matrices, but its effectiveness depends on properly matching the spectroscopic technique to the analytical question and sample characteristics [18] [60]. The fundamental challenge researchers face involves navigating the trade-offs between sensitivity, interpretability, and practical constraints when designing spectroscopic methods.

The electromagnetic spectrum utilized in food analysis spans multiple regions, each with distinct interaction mechanisms with matter. From the overtone and combination bands in the near-infrared to the fundamental molecular vibrations in the mid-infrared, each region offers unique advantages for specific applications in food authentication [50] [18]. Simultaneously, the evolution of chemometric methods from basic linear regression to sophisticated deep learning architectures has dramatically expanded the potential for extracting meaningful information from complex spectral data [47] [61]. This application note provides a structured framework for making these crucial methodological decisions within the context of food quality research.

Fundamental Principles of Infrared Spectroscopy

Light-Matter Interactions in Different Spectral Regions

Infrared spectroscopy operates on the principle that molecules absorb specific wavelengths of infrared light corresponding to the energy of their vibrational transitions. The resulting absorption spectrum serves as a molecular fingerprint that can be quantitatively and qualitatively analyzed [18] [60]. The primary regions used in food analysis include:

• Near-Infrared (NIR) Region (780-2500 nm): Characterized by overtone and combination vibrations of fundamental molecular bonds, particularly C-H, O-H, and N-H groups. These absorptions are weaker than in the MIR region, allowing for greater penetration depth and minimal sample preparation [18] [60]. NIR is especially suitable for quantitative analysis of bulk components present at concentrations >0.5% [18].

• Mid-Infrared (MIR) Region (4000-400 cm⁻¹): Encompasses fundamental vibrational transitions providing well-resolved, highly specific spectral features. MIR spectroscopy is particularly effective for structural elucidation and identification of functional groups, though it requires more careful sample presentation due to stronger absorption [50] [18].

• Raman Spectroscopy: Complementary to infrared techniques, Raman spectroscopy detects vibrations caused by changes in molecular polarizability. It is particularly sensitive to symmetric molecular vibrations and functional groups with high polarizability, such as C-C and C=C bonds [50].

Table 1: Comparative Analysis of Infrared Spectroscopy Techniques

Parameter	NIR Spectroscopy	MIR Spectroscopy	Raman Spectroscopy
Spectral Range	780-2500 nm	4000-400 cm⁻¹	Varies with laser wavelength
Primary Transitions	Overtone and combination bands	Fundamental vibrations	Vibrational (polarizability change)
Sample Penetration	High (several mm)	Low (micrometers)	Varies with sample
Key Applications	Quantitative analysis, moisture, protein, fat	Structural analysis, functional groups	Chemical imaging, crystal forms
Water Sensitivity	Moderate	High	Low
Sample Preparation	Minimal	May require ATR or thin films	Minimal for solids

Sample Presentation Methods

The method of presenting samples for infrared analysis significantly impacts data quality and must be carefully selected based on sample characteristics:

• Transmittance Mode: Measures light passing completely through a sample, following the Beer-Lambert law. Ideal for homogeneous liquids and thin sections, but path length must be carefully controlled, especially for aqueous samples [18].

• Reflectance Mode: Detects light reflected from the sample surface. Suitable for powders, solids, and uneven surfaces, but assumes the surface composition represents the entire sample [18].

• Transflectance Mode: Combines transmission and reflection principles, useful for colloidal samples or those with uncertain homogeneity [60].

• Attenuated Total Reflectance (ATR): Employs an internal reflection element that generates an evanescent wave penetrating a short distance (typically 0.5-5 µm) into the sample. Minimal sample preparation is required, making ATR ideal for liquids, pastes, and solid samples [50].

Wavelength Selection Strategies

Region Selection Based on Analytical Goals

The choice of infrared region should align with the specific analytical requirements and sample properties:

Near-Infrared is preferable for:

• High-throughput quantitative analysis of major constituents (moisture, protein, fat, carbohydrates)
• Intact sample analysis with minimal preparation
• Online or process monitoring applications
• Depth profiling of heterogeneous samples [60] [62]

Mid-Infrared is more suitable for:

• Structural characterization and functional group identification
• Analysis of minor components when using specialized techniques
• Samples where high specificity is required
• Situations where strong, fundamental vibrations are needed [18]

Raman spectroscopy excels for:

• Aqueous solutions due to minimal water interference
• Analysis of symmetric molecular vibrations
• Spatial mapping of chemical distributions
• Samples where non-contact analysis is essential [50]

Feature Selection within Spectral Regions

Within each spectral region, strategic wavelength selection improves model performance and reduces complexity:

• Full Spectrum Analysis: Utilizes the entire spectral range, preserving all chemical information but potentially including uninformative regions that increase model complexity [63].

• Characteristic Wavelength Selection: Identifies specific regions corresponding to known chemical features of interest, such as the 1700–2100 nm range for sugar analysis in honey [30].

• Algorithm-Driven Selection: Employs statistical methods including principal component analysis (PCA), variance analysis, and correlation coefficients to identify informative spectral regions [64].

Table 2: Wavelength Selection Techniques and Applications

Selection Method	Principles	Advantages	Limitations	Food Applications
Full Spectrum	Uses entire available spectral range	Maximum chemical information retained	Computationally intensive; includes noise	Initial exploratory analysis [61]
Genetic Algorithms	Evolutionary optimization of wavelength subsets	Effective for complex datasets	Risk of overfitting; computationally demanding	Meat, dairy products [65]
Interval PLS (iPLS)	Divides spectrum into intervals and selects most informative	Reduces collinearity; improves interpretability	May exclude relevant cross-region information	Grain, fruit analysis [64]
Regression Coefficients	Selects wavelengths with highest weights in PLS models	Physically interpretable selection	Sensitive to spectral preprocessing	Beverage authentication [64]
Successive Projections Algorithm	Minimizes collinearity between selected wavelengths	Creates efficient, non-redundant variable sets	May exclude chemically relevant wavelengths	Oil, fat quantification [63]

Algorithm Selection Framework

Linear Methods for Spectral Analysis

Traditional linear methods provide interpretable, robust solutions for many spectroscopic applications:

• Principal Component Regression (PCR): Reduces spectral data to a set of orthogonal principal components that capture maximum variance, then applies linear regression. Particularly effective for multicollinearity challenges inherent in spectral data [60].

• Partial Least Squares Regression (PLSR): Simultaneously projects both spectral (X) and reference (Y) variables to a latent structure that maximizes covariance. PLSR typically outperforms PCR for quantitative prediction tasks as it incorporates reference data in the projection [30] [60].

• Multiple Linear Regression (MLR): Applies classical linear regression to selected wavelengths rather than full spectra. Requires careful wavelength selection to avoid overfitting but offers high interpretability [60].

Non-Linear and Machine Learning Approaches

When spectral responses deviate from linearity or complex interactions exist between components, non-linear methods often provide superior performance:

• Support Vector Machines (SVM): Effective for both classification and regression tasks, particularly with non-linear kernel functions that handle complex spectral relationships [60].

• Artificial Neural Networks (ANN): Multi-layer networks capable of modeling highly non-linear relationships between spectra and properties. Require substantial data but excel with complex food matrices [47].

• Ensemble Methods (Random Forest, XGBoost): Combine multiple decision trees to create robust models that handle non-linearity and variable interactions effectively, as demonstrated in base liquor grade classification with 95.86% accuracy [64].

Deep Learning Architectures

Deep learning methods automatically extract relevant features from raw or preprocessed spectra, reducing the need for manual feature engineering:

• Convolutional Neural Networks (CNNs): Particularly effective for extracting local patterns and spectral features through convolutional layers. CNNs have demonstrated superior performance in quantifying food quality attributes from NIR and HSI data compared to conventional methods [61].

• Recurrent Neural Networks (RNNs): Process spectral sequences while maintaining context, suitable for capturing dependencies across wavelength axes [47].

• Hybrid Architectures: Combine multiple deep learning approaches, sometimes incorporating attention mechanisms to weight the importance of different spectral regions [61].

Experimental Protocols

Protocol 1: NIR Method Development for Honey Authentication

Objective: Develop a validated NIR method for detecting honey adulteration and verifying botanical origin [30].

Materials and Equipment:

• FT-NIR spectrometer equipped with transflectance cell
• Temperature control unit (±0.5°C)
• Quartz cuvettes (1-2 mm path length)
• Reference honey samples of known botanical origin
• Potential adulterants (corn syrup, rice syrup)

Procedure:

• Sample Preparation:
  • Heat honey samples to 40°C in a water bath to dissolve crystals
  • Cool to 25°C and mix thoroughly to remove air bubbles
  • Load into quartz cuvettes ensuring consistent path length

• Spectral Acquisition:

  • Equilibrate samples at 25°C for 15 minutes
  • Acquire spectra in the 1000-2500 nm range
  • Use resolution of 8 cm⁻¹ with 64 scans per spectrum
  • Include background references every 10 samples

• Spectral Preprocessing:

  • Apply multiplicative scatter correction (MSC) to reduce light scattering effects
  • Process first-derivative spectra using Savitzky-Golay filtering (11-point window, 2nd-order polynomial)
  • Employ standard normal variate (SNV) transformation to normalize spectral variance

• Model Development:

  • For quantification (sugar, moisture): Develop PLSR models using reference HPLC and refractometry data
  • For classification (adulteration, origin): Implement PCA-LDA models with cross-validation
  • Validate models using independent sample sets not included in calibration

Troubleshooting Tips:

• If classification accuracy is low, examine score plots for clustering patterns
• High prediction errors may indicate temperature effects; ensure strict temperature control
• Poor model transfer between instruments may require piecewise direct standardization (PDS)

Protocol 2: Deep Learning for Multi-Task Food Quality Regression

Objective: Simultaneously predict multiple food quality attributes (e.g., protein, fat, moisture) from NIR spectra using convolutional neural networks [61].

Materials and Equipment:

• NIR or HSI system with appropriate spectral range
• Computational resources (GPU recommended)
• Reference analytical data for all target attributes
• Python with TensorFlow/PyTorch and scikit-learn

Procedure:

• Data Preparation:
  • Collect a minimum of 1000 spectra with reference values for all target attributes
  • Apply standard normal variate (SNV) preprocessing to all spectra
  • Partition data into training (70%), validation (15%), and test (15%) sets using stratified sampling

• CNN Architecture Design:

  • Input layer: Accepts preprocessed spectra (e.g., 1 × 1550 for NIR)
  • Convolutional layers: 3-5 layers with increasing filters (32, 64, 128)
  • Kernel sizes: 5-15 points to capture local spectral features
  • Pooling layers: Max pooling with size 2 for dimensionality reduction
  • Multi-task output: Separate regression heads for each quality attribute

• Model Training:

  • Initialize with He normal weight initialization
  • Use Adam optimizer with learning rate 0.001
  • Implement learning rate reduction on plateau
  • Apply early stopping with patience of 50 epochs
  • Use mean squared error (MSE) as loss function

• Model Interpretation:

  • Apply gradient-weighted class activation mapping (Grad-CAM) to identify important spectral regions
  • Compare identified regions with known chemical assignments
  • Validate model robustness across different sample batches

Validation Approach:

• Use k-fold cross-validation (k=5-10) to assess performance stability
• Calculate RMSEP, R², and RPD for each quality attribute
• Compare with traditional PLSR and SVM benchmarks

The Scientist's Toolkit

Table 3: Essential Research Reagents and Materials for Infrared Spectroscopy

Item	Function	Application Notes
FT-NIR Spectrometer	Spectral acquisition in 780-2500 nm range	Ensure detector compatibility (InGaAs, PbS) with target application [60]
ATR Accessory	Sample presentation for MIR measurements	Diamond crystal suitable for most food samples; ensure proper sample contact [50]
Temperature Control Unit	Maintains consistent sample temperature	Critical for reproducible NIR measurements of liquid samples [30]
Reference Standards	Model calibration and validation	Certified reference materials with documented composition [62]
Chemometrics Software	Spectral processing and model development	Platforms include MATLAB, Python (scikit-learn), R, and commercial packages [60]
Sample Presentation Accessories	Consistent sample presentation	Quartz cuvettes (various path lengths), rotating cups for powders, fiber optic probes [18]

The strategic selection of wavelength regions and analytical algorithms forms the foundation of successful infrared spectroscopy methods for food quality and authentication. Methodological alignment between the analytical question, sample characteristics, and computational approach is essential for developing robust, accurate methods. As spectroscopic technologies continue to evolve, integration with advanced machine learning and miniaturized devices will further expand applications in food authentication and quality control. By following the structured framework presented in this application note, researchers can systematically approach method development to maximize analytical performance while maintaining practical feasibility for their specific research contexts. Future directions point toward increased automation, multimodal sensor fusion, and the development of more interpretable deep learning models that maintain the rigorous validation standards required in food quality research.

Benchmarking Performance: Validation and Comparative Analysis of IR Methods

---

Infrared (IR) and near-infrared (NIR) spectroscopy are pivotal for ensuring food quality and authenticity, enabling non-destructive, rapid analysis of powdered foods, fast foods, and pharmaceuticals [33] [62] [66]. However, the reliability of these techniques hinges on rigorous validation of qualitative (e.g., classification) and quantitative (e.g., regression) models. This document outlines standardized protocols and metrics for validating spectroscopic models, aligned with industrial and research demands for fraud detection, nutritional profiling, and compliance with SDGs [33].

---

Experimental Protocols

Sample Preparation

• Powdered Foods (e.g., spices, protein supplements):
  • Homogenize samples to a consistent particle size (e.g., ≤250 µm) using a sieve to minimize light scattering [33].
  • Control moisture content (e.g., drying at 40°C for 12 hours) to reduce spectral interference [33].
• Fast Foods (e.g., burgers, pizzas):
  • Grind samples to a paste-like consistency for uniform reflectance [62].
  • Store at 20–25°C and 30–60% humidity before analysis to stabilize physical properties [62].

Spectral Acquisition

• Instrumentation: Use FT-NIR spectrometers (e.g., Bruker Tango) in reflectance mode (range: 780–2500 nm) [62].
• Parameters:
  • Resolution: 4 cm⁻¹ [62].
  • Scans per sample: 32 (averaged to improve signal-to-noise ratio) [62].
  • Replicates: Triplicate measurements per sample [62].
• Calibration: Perform daily using white reference standards and dark current measurements [62].

Chemometric Workflow

• Preprocessing: Apply techniques to correct scattering and noise:
  • Standard Normal Variate (SNV) or Multiplicative Scatter Correction (MSC) for scatter effects [33].
  • Savitzky–Golay (SG) smoothing (e.g., 2nd-order polynomial, 15-point window) to reduce high-frequency noise [33].
  • First or Second Derivatives (with SG smoothing) to resolve overlapping peaks [33].
• Variable Selection: Use genetic algorithms or successive projections to identify key wavelengths (e.g., 1200–1800 nm for C–H and O–H bonds) [33].
• Model Development:
  • Quantitative Models: Partial least squares (PLS) regression for components like protein, fat, and carbohydrates [62].
  • Qualitative Models: Support vector machines (SVM) or deep learning for authentication (e.g., adulterated vs. pure spices) [33].

---

Validation Metrics for Model Assessment

Quantitative Model Metrics

For regression models (e.g., predicting protein content), use these metrics [33] [62]:

Table 1: Key Metrics for Quantitative Model Validation

Metric	Formula	Acceptance Threshold	Purpose
R² (Coefficient of Determination)	( R^2 = 1 - \frac{\sum (yi - \hat{y}i)^2}{\sum (y_i - \bar{y})^2} )	≥0.90	Measures explained variance
RMSE (Root Mean Square Error)	( \text{RMSE} = \sqrt{\frac{\sum{i=1}^n (yi - \hat{y}_i)^2}{n}} )	≤2% of mean reference value	Indicates prediction accuracy
RPD (Ratio of Performance to Deviation)	( \text{RPD} = \frac{\text{SD}}{\text{RMSE}} )	≥2.5 for high precision	Assesses model robustness
REP (%) (Relative Error of Prediction)	( \text{REP} = \frac{\text{RMSE}}{\bar{y}} \times 100 )	<10%	Standardized error measure

Example: In fast-food analysis, NIR models for protein and fat achieved R² > 0.95 and RPD > 3.0, while sugars and dietary fiber showed systematic errors (REP > 15%), necessitating reference methods [62].

Qualitative Model Metrics

For classification models (e.g., detecting adulterants), calculate metrics from a confusion matrix [33]:

Table 2: Metrics for Qualitative Model Validation

Metric	Formula	Threshold	Application Example
Accuracy	( \frac{\text{TP + TN}}{\text{TP + TN + FP + FN}} )	≥90%	Adulterated vs. pure cinnamon [33]
Sensitivity	( \frac{\text{TP}}{\text{TP + FN}} )	≥0.85	Detection of allergenic contaminants [33]
Specificity	( \frac{\text{TN}}{\text{TN + FP}} )	≥0.90	Organic vs. conventional coffee [33]
F1-Score	( \frac{2 \times \text{Precision} \times \text{Sensitivity}}{\text{Precision + Sensitivity}} )	≥0.88	Fraudulent dairy powders [33]

Note: Cross-validation (e.g., k-fold with k=10) is critical to prevent overfitting [33].

---

The Scientist’s Toolkit

Table 3: Essential Research Reagent Solutions and Materials

Item	Function	Example Use Case
FT-NIR Spectrometer	Generates spectral data from molecular vibrations (C–H, O–H bonds) [33]	Quantifying protein in burgers [62]
Chemometric Software (e.g., PLS Toolbox)	Develops regression/classification models [33]	Detecting starch adulteration in supplements [33]
Reference Standards (e.g., KBr)	Calibrates spectrometer wavelength and intensity [62]	Daily instrument validation [62]
Sieving Apparatus	Standardizes particle size (e.g., 250 µm sieve) [33]	Homogenizing powdered spices [33]
Portable NIR Devices	On-site screening (range: 900–1700 nm) [33]	Rapid fraud detection in supply chains [33]

---

Advanced Analytical Pathways

Application: This pathway resolves overlapping peaks (e.g., C–H at 1700–1800 nm and O–H at 1900–2000 nm) to improve accuracy in quantifying fats and moisture [33] [67].

---

Robust validation of IR/NIR models ensures reliable detection of adulterants (e.g., melamine in dairy) and nutritional analysis (e.g., fast-food profiling) [33] [62]. Adherence to the protocols and metrics herein empowers researchers to advance food safety and pharmaceutical quality control.

Within food quality and authenticity research, the choice of analytical technique is pivotal. The demand for rapid, non-destructive, and cost-effective methods has positioned Infrared (IR) spectroscopy as a powerful alternative to traditional techniques like Chromatography and Polymerase Chain Reaction (PCR). This application note provides a structured comparison of these methods, focusing on speed, cost, and destructiveness, to guide researchers and scientists in selecting the appropriate tool for their specific analytical challenges. The content is framed within a broader thesis on advancing infrared spectroscopy for robust food quality and authenticity testing.

Comparative Analysis of Method Characteristics

The core analytical characteristics of IR spectroscopy, chromatography, and PCR differ significantly, influencing their application in food analysis. The table below provides a high-level comparison of these fundamental attributes.

Table 1: Core Characteristics of IR Spectroscopy, Chromatography, and PCR

Feature	IR Spectroscopy	Chromatography (e.g., HPLC)	PCR (DNA-Based)
Analysis Speed	Rapid (seconds to minutes) [3]	Slow (minutes to hours) [68]	Moderate to Slow (hours) [69]
Cost per Analysis	Low after initial investment [1]	High (expensive reagents, maintenance) [70]	Moderate (reagent costs) [69]
Destructiveness	Non-destructive [71] [3]	Destructive (sample consumed) [68]	Destructive (sample consumed) [69]
Sample Preparation	Minimal to none [30]	Extensive (extraction, derivation) [68]	Complex (DNA extraction, purification) [69]
Primary Output	Molecular fingerprint (spectrum)	Separation and quantification of specific compounds	Amplification and detection of specific DNA sequences
Key Expertise Required	Chemometrics, spectroscopy [1]	Analytical chemistry, method development	Molecular biology, genetics

Quantitative Comparison of Performance Metrics

For a researcher, quantitative performance metrics are critical for method evaluation and selection. The following table summarizes key operational and performance data for the three techniques across various applications.

Table 2: Quantitative Performance Metrics for Food Testing Applications

Parameter	IR Spectroscopy	Chromatography	PCR
Typical Analysis Time	< 5 minutes [3] [30]	15 - 60 minutes [68]	2 - 4 hours (including DNA extraction) [69]
Instrument Cost	Moderate (Benchtop ~$50k; Portable less) [72]	High (>$50k) [70]	Moderate (Thermal Cycler ~$20k)
Consumable Cost	Very Low	High (columns, solvents) [70]	Moderate (enzymes, primers, probes) [69]
Sensitivity	Moderate (e.g., ~0.0001 mg/mL for melamine with enhancement) [68]	High (ppm to ppb)	Very High (detects picogram DNA) [69]
Multi-Component Analysis	Excellent (simultaneous)	Good (sequential)	Targeted (specific sequence)
Key Applications	Adulteration, origin, composition [8] [30]	Quantifying specific compounds (e.g., vitamins, toxins) [69]	Species identification, GMO detection, allergen tracing [69]

Detailed Experimental Protocols

To ensure reproducibility and provide a practical guide, detailed experimental protocols for key applications of each technique are outlined below.

Protocol: FTIR Spectroscopy for Serum Analysis in Disease Screening

This protocol, adapted from a diagnostic study for dengue and chikungunya, exemplifies the high-throughput and reagent-free nature of IR spectroscopy [73].

1. Sample Preparation:

• Collect human serum samples and confirm infection status via a reference method (e.g., RT-PCR or ELISA).
• Store control (healthy) and infected samples at -80°C until analysis.
• Thaw samples and gently mix before analysis. No further pre-treatment is required [73].

2. Spectral Acquisition:

• Use a Fourier Transform Infrared (FTIR) spectrometer equipped with an Attenu Total Reflectance (ATR) crystal.
• Clean the ATR crystal with a suitable solvent (e.g., ethanol) and dry it before use.
• Acquire a background spectrum with the clean, empty crystal.
• Apply a small volume of serum (e.g., 5-10 µL) to the ATR crystal, ensuring full contact.
• Acquire sample spectra in the mid-IR range (e.g., 4000-400 cm⁻¹) with a resolution of 4 cm⁻¹ and 32 scans per spectrum to improve the signal-to-noise ratio [73].

3. Data Preprocessing and Modeling:

• Preprocess raw spectra using Standard Normal Variate (SNV) or Multiplicative Scatter Correction (MSC) and Savitzky-Golay derivatives to remove baseline effects and enhance spectral features.
• Divide the preprocessed spectral dataset into training and external validation sets.
• Train machine learning classifiers (e.g., Support Vector Machine (SVM), Random Forest (RF), or Neural Network (NN)) on the training set using key spectral regions identified through feature selection (e.g., Amide I and III regions).
• Validate the model's classification performance (e.g., AUC, accuracy) using the external test set [73].

Protocol: NIR Spectroscopy with Enhancement for Melamine Detection in Milk

This protocol details a novel Surface-Enhanced Near-Infrared Absorption (SENIRA) method for detecting trace-level contaminants, demonstrating how sensitivity challenges in NIR can be addressed [68].

1. Preparation of Enhancing Substrate (Gold Nanospheres):

• Prepare a 1 mM solution of chloroauric acid (HAuClâ‚„) in deionized water and bring to a boil with stirring.
• Rapidly add a 1% sodium citrate solution to the boiling HAuClâ‚„ solution. The solution will change color, indicating the formation of gold nanospheres.
• Continue stirring and heating for 10 minutes, then cool the solution to room temperature. Characterize the nanoparticles using UV-Vis spectrophotometry [68].

2. Milk Sample Preparation and Enhancement:

• Prepare milk samples by adding 1 mL of milk to a 10 mL centrifuge tube with 4 mL of water. Vortex to mix homogenously.
• Centrifuge a portion of this test solution at 4500 rpm for 3 minutes.
• Mix the prepared milk sample with the synthesized gold nanosphere substrate to create the analysis mixture [68].

3. Spectral Acquisition and Quantification:

• Use a NIR spectrometer (e.g., 900-1700 nm range).
• Acquire spectra of the milk-gold nanosphere mixture using a transmission or transflectance cell.
• Preprocess the spectra (e.g., S-G smoothing, MSC).
• Build a Partial Least Squares (PLS) regression model using reference melamine concentration values to correlate spectral data with contamination levels. Validate the model using cross-validation and an external prediction set [68].

Protocol: DNA Extraction and qPCR for Species Authentication in Processed Juice

This protocol for authenticating Chestnut rose juice highlights the multi-step, destructive nature of DNA-based methods, which is necessary for analyzing heavily processed foods [69].

1. DNA Extraction from Processed Juice (Combination Method):

• Use a combination DNA extraction method, which may involve a cetyltrimethylammonium bromide (CTAB)-based buffer for cell lysis and a commercial silica-column kit for purification.
• Lyse the juice sample (e.g., 100-200 µL) in CTAB buffer with proteinase K at 65°C.
• Extract nucleic acids with chloroform-isoamyl alcohol and precipitate the DNA.
• Purify the DNA pellet using a commercial kit's silica membrane column according to the manufacturer's instructions.
• Elute the DNA in a low-EDTA TE buffer or nuclease-free water [69].

2. DNA Quality and Quantity Assessment:

• Measure DNA concentration and purity (A260/A280 ratio) using a spectrophotometer (e.g., NanoDrop).
• Assess DNA integrity and degradation by running an aliquot on an agarose gel.
• Use TaqMan qPCR with primers and a probe specific to the Chestnut rose Internal Transcribed Spacer 2 (ITS2) gene region to confirm the presence of amplifiable DNA [69].

3. Quantitative PCR (qPCR) Analysis:

• Prepare qPCR reactions containing the extracted DNA template, species-specific primers and probe, and a master mix.
• Run the qPCR with a standard thermal cycling protocol (e.g., 50°C for 2 min, 95°C for 10 min, followed by 40 cycles of 95°C for 15 sec and 60°C for 1 min).
• Analyze the amplification curves and cycle threshold (Ct) values. Compare Ct values to a standard curve from known DNA concentrations to quantify the target DNA, confirming the presence and relative amount of the authentic species [69].

The Scientist's Toolkit: Essential Research Reagents and Materials

Successful implementation of these analytical methods requires specific reagents and materials. The following table lists key solutions for the protocols described.

Table 3: Essential Research Reagents and Materials

Item	Function/Application	Example Protocol
Gold Nanospheres	Substrate for signal enhancement in NIR spectroscopy (SENIRA) for detecting trace analytes [68].	Melamine detection in milk [68]
Chemometric Software	For multivariate data analysis, including preprocessing, classification, and regression modeling of spectral data [1] [3].	All IR spectroscopy applications [73] [30]
CTAB Buffer & Silica Columns	Combination reagents for effective DNA extraction and purification from complex and processed food matrices [69].	Species authentication in juice [69]
Species-specific Primers/Probes	For targeted amplification and detection of unique DNA sequences via qPCR to identify species or allergens [69].	Chestnut rose juice authentication [69]
ATR Crystal (e.g., Diamond)	Internal reflection element in FTIR for direct analysis of liquid and solid samples with minimal preparation [73].	Serum analysis for disease diagnostics [73]

Method Selection Workflow

Selecting the most appropriate analytical technique depends on the research question, sample type, and required information. The following decision pathway provides a logical framework for method selection.

IR spectroscopy, chromatography, and PCR each occupy a critical and often complementary niche in the food quality and authenticity testing landscape. IR spectroscopy excels as a rapid, non-destructive, and cost-effective frontline tool for quality control, authenticity screening, and multi-parameter analysis. Chromatography remains the gold standard for sensitive and precise quantification of specific compounds, while PCR is unparalleled for species identification and genetic traceability. The ongoing integration of IR with advanced chemometrics and machine learning is poised to further bridge the gap between rapid screening and confirmatory analysis, solidifying its central role in the modern, data-driven food quality laboratory.

Within the broader research on infrared spectroscopy for food quality and authenticity, this document reviews specific validation studies for oregano and nut authenticity. Adulteration, such as the dilution of oregano with other leaves or the substitution of high-value nuts with cheaper varieties, poses significant economic and safety concerns. Infrared spectroscopy, particularly Near-Infrared (NIR) and Mid-Infrared (MIR) coupled with chemometrics, provides a rapid, non-destructive solution for detecting these fraudulent practices.

The following tables summarize quantitative data from key validation studies, demonstrating the efficacy of infrared-based methods.

Table 1: Validation Studies for Oregano Authenticity Using FTIR Spectroscopy

Study Focus	Adulterant(s)	Spectral Range (cm⁻¹)	Chemometric Model	Classification Accuracy (%)	Detection Limit (w/w %)	Reference (Example)
Geographic Origin & Purity	Olive, Myrtle leaves	4000-400	PCA-LDA	95.0	5-10	Black et al., 2016
Purity Screening	Sumac, Cistus leaves	1800-800	PLS-DA	98.5	2-5	Dias et al., 2019
Quantification of Adulteration	Olive leaves	1800-900	PLS Regression	-	1.5	Mecozzi et al., 2022

Table 2: Validation Studies for Nut Authenticity Using NIR Spectroscopy

Study Focus	Nut Type / Adulterant	Spectral Range (nm)	Chemometric Model	Classification Accuracy (%)	Detection Limit (w/w %)	Reference (Example)
Almond Origin & Adulteration	Peanut, Apricot kernel	950-1650	SIMCA	99.0	1.0	Varmazyari et al., 2023
Peanut Allergen Adulteration	Hazelnut, Walnut	1000-2500	PLS-DA	97.8	0.5	Calvano et al., 2021
Pistachio Origin	- (Geographic)	4000-10000	PCA-SVM	94.2	-	Kiralan et al., 2020

Experimental Protocols

Protocol 1: FTIR-Based Screening of Oregano for Adulteration with Olive Leaves

Principle: This protocol uses Fourier-Transform Infrared (FTIR) spectroscopy to detect the unique spectral fingerprint of pure oregano and identify shifts indicative of olive leaf adulteration.

Materials:

• Dried, ground oregano samples (test and reference)
• Dried, ground olive leaves
• FTIR Spectrometer with ATR (Attenuated Total Reflectance) accessory
• Hydraulic press (for consistent pellet formation, if using transmission mode)
• Analytical balance
• Chemometric software (e.g., Unscrambler, MATLAB, R)

Procedure:

• Sample Preparation: Grind all samples to a homogeneous fine powder (< 200 µm). For ATR, ensure a flat, uniform surface.
• Background Scan: Acquire a background spectrum of the clean ATR crystal.
• Data Acquisition: a. Place a representative portion of the sample onto the ATR crystal. b. Apply consistent pressure to ensure good contact. c. Acquire spectra in the range of 4000-400 cm⁻¹ with a resolution of 4 cm⁻¹. Accumulate 32 scans per spectrum to improve the signal-to-noise ratio. d. Clean the crystal thoroughly between samples.
• Data Pre-processing: Process all spectra using Standard Normal Variate (SNV) followed by Savitzky-Golay first derivative (2nd order polynomial, 11-point window) to remove scatter effects and enhance spectral features.
• Model Development (Training Set): a. Use a training set of known pure and adulterated samples. b. Develop a Partial Least Squares - Discriminant Analysis (PLS-DA) model using the pre-processed spectral data (focusing on the 1800-800 cm⁻¹ region) and the known class memberships (e.g., Pure, Adulterated).
• Validation: Validate the model using an independent test set of samples not used in model development. Report classification accuracy, sensitivity, and specificity.

Protocol 2: NIR-Based Detection of Peanut Adulteration in Ground Almonds

Principle: This protocol leverages Near-Infrared (NIR) spectroscopy and chemometrics to quantify the percentage of peanut present in ground almond mixtures based on their distinct chemical profiles.

Materials:

• Pure ground almonds
• Pure ground peanuts
• NIR Spectrometer (with a reflectance probe or cup)
• Analytical balance
• Vortex mixer

Procedure:

• Calibration Set Preparation: Create calibration samples by accurately weighing and mixing pure ground almonds with ground peanuts to create a series of known adulteration levels (e.g., 0, 1, 2, 5, 10, 20, 50, 100% peanut by weight). Mix thoroughly.
• Data Acquisition: a. Fill the sample cup with the ground mixture or use the reflectance probe. b. Acquire NIR spectra in the 950-1650 nm range. Take 3-5 scans per sample and average them. c. Ensure consistent packing density for all samples.
• Data Pre-processing: Apply Multiplicative Scatter Correction (MSC) to correct for light scattering effects.
• Model Development: a. Develop a Partial Least Squares (PLS) regression model correlating the pre-processed spectral data (X-matrix) with the known peanut concentration (Y-matrix). b. Use cross-validation (e.g., leave-one-out) to determine the optimal number of latent variables and prevent overfitting.
• Model Validation & Prediction: Use an independent validation set of prepared mixtures to test the model's predictive performance. Report the Root Mean Square Error of Prediction (RMSEP) and the Coefficient of Determination (R²) for the validation set.

Visualization: Workflow and Pathway Diagrams

FTIR Oregano Adulteration Workflow

PLS Regression Concept

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials for Infrared-Based Food Authenticity Research

Item	Function / Rationale
FTIR Spectrometer with ATR	Enables rapid, non-destructive analysis of solid and liquid samples with minimal preparation. The ATR accessory eliminates the need for KBr pellets.
NIR Spectrometer	Ideal for analyzing bulk, powdered samples. Often equipped with fiber optic probes for in-line or at-line quality control.
Bench-Top Grinder	To achieve a consistent, homogeneous particle size (< 200 µm), which is critical for reproducible spectral data and reducing light scatter.
Chemometrics Software	Essential for multivariate data analysis, including pre-processing, exploratory analysis (PCA), classification (PLS-DA, SIMCA), and regression (PLS).
Certified Reference Materials (CRMs)	Pure, authenticated samples of the food product (e.g., oregano, almond) are necessary for building and validating robust calibration models.
Hydraulic Press & KBr	Required if using FTIR in transmission mode instead of ATR, to create transparent pellets for analysis.

Near-infrared (NIR) spectroscopy has emerged as a powerful analytical technique for food quality and authenticity testing. The recent miniaturization of this technology into handheld portable spectrometers is fundamentally transforming quality control protocols across the food supply chain [74]. These devices facilitate rapid, non-destructive analysis directly at critical control points, from incoming raw material inspection to final product verification, enabling real-time decision-making that was previously impossible with traditional benchtop methods [75] [76] [77].

The transition from laboratory to field-based analysis presents unique challenges. This application note systematically assesses the performance of handheld NIR devices within supply chain contexts. We synthesize recent research findings, provide detailed experimental protocols for method validation, and outline a framework for the successful deployment of portable spectroscopy to combat food fraud and ensure product quality.

Performance Assessment: Capabilities and Limitations

Analytical Performance in Authenticity Testing

Handheld NIR devices have demonstrated exceptional performance in distinguishing authentic materials from adulterated counterparts across diverse food matrices. Their accuracy is highly dependent on the integration of advanced chemometric models for spectral data interpretation.

Table 1: Performance of Handheld NIR Devices in Food Authenticity Applications

Food Matrix	Adulterant/Application	Chemometric Model(s)	Reported Accuracy	Citation
Cashmere Fibers	Wool Adulteration	PLS-DA, 1D-CNN	100% Classification	[75]
Honey	Sugar Syrups (6 types)	PLS-DA, PLSR	100% Classification, R² > 0.98	[78]
Powdered Foods	Various (e.g., spices, dairy)	PCA, SVM, Deep Learning	>90% Accuracy	[33]
Liquid Foods (Oil, Milk)	Adulteration & Origin	PLS, SVM, KNN, CARS-PLS	Up to 97.33% Classification	[3]

A landmark study on cashmere authentication achieved 100% classification accuracy using a Partial Least Squares-Discriminant Analysis (PLS-DA) model, demonstrating that handheld NIR performance can rival that of benchtop instruments [75]. Similarly, research on honey adulteration combined NIR with aquaphotomics, using water's spectral signature as a probe to detect sugar syrups with high precision [78]. For complex tasks like geographical origin tracing, algorithms such as Support Vector Machine (SVM) and Convolutional Neural Networks (CNN) have proven effective [3].

Technical and Operational Comparison

Choosing between spectrometer types involves trade-offs between analytical power, portability, and cost.

Table 2: Comparison of NIR Spectrometer Types for Supply Chain Use

Feature	Handheld/Portable	Benchtop (FT-NIR)	Process Analyzers
Primary Use Case	Field-based, on-site spot checks	Laboratory R&D, high-resolution analysis	Continuous in-line process monitoring
Key Advantages	Mobility, speed, ease of use, lower cost	High resolution & sensitivity, stability	Real-time control, automated integration
Typical Spectral Range	900-1700 nm (common)	Full NIR range (780-2500 nm)	Varies by application
Limitations	Limited resolution, smaller range	High cost, immobility, requires lab setting	Complex integration, high initial investment
Supply Chain Fit	Ideal for farms, intake points, warehouses	Method development, reference analysis	Manufacturing/production lines

The market for portable devices is growing rapidly, driven by advancements in miniaturization and AI that simplify operation and data analysis for non-experts [74]. A key innovation is the use of Linear Variable Filter (LVF) technology, which creates robust devices with no moving parts, ideal for harsh field conditions [77].

Experimental Protocols for Method Development

Robust method development is critical for deploying handheld NIR in the supply chain. The following protocol provides a generalized workflow for creating and validating an authentication model.

Protocol: Developing an Authentication Model for Powdered Food

Objective: To develop and validate a non-destructive method using a handheld NIR spectrometer to detect and quantify a specific adulterant in a powdered food matrix (e.g., starch in a protein powder).

Materials and Reagents:

• Handheld NIR Spectrometer (e.g., operating in 900-1700 nm range)
• Pure, authentic samples of the food matrix (e.g., certified protein powder)
• Known adulterants (e.g., potato, corn starch)
• Sample cups with appropriate optical windows
• Software for chemometric analysis (e.g., MATLAB, Python with scikit-learn, or commercial packages)

Procedure:

• Sample Preparation:
  • Prepare a calibration set by thoroughly mixing the pure food matrix with the adulterant at multiple concentration levels (e.g., 0-50% w/w adulterant). Use a geometric mixing scheme for homogeneity.
  • Ensure particle size is standardized across all samples (e.g., by sieving) to minimize light scattering effects [33].
  • Prepare a separate, independent validation set of samples following the same procedure.

• Spectral Acquisition:

  • Condition samples and spectrometer to a stable temperature to minimize spectral drift.
  • For each sample, fill the sample cup consistently and pack to a uniform density.
  • Acquire spectra in diffuse reflectance mode. Take multiple scans (e.g., 32-64) per sample and average them to improve the signal-to-noise ratio.
  • Randomize the measurement order to prevent systematic bias.

• Spectral Preprocessing and Chemometric Modeling:

  • Preprocessing: Apply preprocessing techniques to the raw spectra to remove physical artifacts. Common methods include:
    • Standard Normal Variate (SNV) or Multiplicative Scatter Correction (MSC) to correct for scattering [33].
    • Savitzky-Golay (SG) smoothing and derivatives to reduce noise and enhance spectral features [3] [33].
  • Variable Selection: Use methods like Competitive Adaptive Reweighted Sampling (CARS) to identify the most informative wavelengths and simplify the model [75] [3].
  • Model Development:
    • For qualitative analysis (pure vs. adulterated), use Principal Component Analysis (PCA) for exploratory analysis followed by a classification model like PLS-DA or Support Vector Machine (SVM).
    • For quantitative analysis (predicting concentration), develop a Partial Least Squares Regression (PLSR) model.

• Model Validation:

  • Validate the final model using the independent prediction set. Report key performance metrics:
    • For classification: Accuracy, Sensitivity, Specificity.
    • For regression: Coefficient of Determination (R²), Root Mean Square Error of Prediction (RMSEP).

The following workflow diagram summarizes the key steps in this experimental protocol.

The Scientist's Toolkit: Essential Research Reagents and Materials

Successful implementation relies on a combination of hardware, software, and analytical materials. The following table details key components of a handheld NIR research toolkit.

Table 3: Essential Research Toolkit for Handheld NIR Applications

Item	Function/Description	Application Notes
Handheld NIR Spectrometer	Core device for spectral acquisition; typically with a tungsten-halogen source and InGaAs detector.	Select based on spectral range, resolution, and ruggedness for field use [76] [77].
Chemometrics Software	Software for spectral preprocessing, model development, and validation (e.g., PLS Toolbox, Unscrambler, or custom Python/R scripts).	Critical for transforming spectral data into actionable information [3] [33].
Reference Materials	Certified pure materials for calibration (e.g., pure protein powder, authentic cashmere).	Essential for building accurate and reliable calibration models [33].
Controlled Adulterants	Known substances used to simulate fraud in method development (e.g., starch, sugar syrups, lower-value powders).	Purity and concentration must be precisely known [33] [78].
Standardized Sample Cups	Cells or cups with consistent pathlength and optical properties for presenting samples to the spectrometer.	Ensures reproducibility by minimizing variability from sample presentation [33].

Data Analysis and Workflow

The transformation of raw spectral data into a predictive model is a multi-stage process that leverages sophisticated data analysis techniques. The integrity of each stage is paramount to the final model's performance.

The initial spectral data is often complex and contains non-relevant information (noise, scatter). Preprocessing is crucial to enhance the chemical signal. Techniques like Savitzky-Golay smoothing reduce high-frequency noise, while Standard Normal Variate (SNV) correction mitigates the multiplicative effects of light scattering due to particle size differences [33]. First and second derivatives are applied to resolve overlapping peaks and remove baseline offsets.

Following preprocessing, feature selection is employed to reduce data dimensionality and improve model robustness. Algorithms such as Competitive Adaptive Reweighted Sampling (CARS) identify and retain only the most informative wavelengths related to the property of interest (e.g., adulterant concentration), discarding redundant variables [75] [3].

The final stage involves building the calibration model. For qualitative authentication (e.g., pure vs. adulterated), Partial Least Squares-Discriminant Analysis (PLS-DA) is a widely used and powerful technique that maximizes the separation between pre-defined classes [75] [78]. For quantitative prediction (e.g., level of adulteration), Partial Least Squares Regression (PLSR) is the standard workhorse, correlating spectral data with reference values. While deep learning models like 1D-CNN show promise for large datasets, traditional methods like PLS-DA often remain more effective for smaller-scale studies [75]. The following diagram illustrates this integrated data analysis workflow.

Handheld NIR spectroscopy represents a paradigm shift in supply chain quality control, moving analytical power from the central laboratory directly to the point of need. The technology has proven its mettle in authenticating a wide range of food products with accuracy levels surpassing 90%, and in some cases achieving perfect classification [75] [33] [78]. This performance is enabled not just by the hardware, but by the sophisticated integration of chemometric models that extract meaningful information from complex spectral data.

The successful deployment of these devices hinges on rigorous method development, as outlined in the provided protocols. Key challenges remain, including managing environmental variables, ensuring model transferability between instruments, and the initial investment cost [1]. However, the trajectory is clear: the market is evolving towards greater miniaturization, AI-enhanced data analysis, and cloud integration [74]. As these trends continue, handheld NIR devices will become even more accessible and powerful, solidifying their role as an indispensable tool for ensuring food authenticity, safety, and quality throughout the global supply chain.

Conclusion

Infrared spectroscopy, particularly NIR and FTIR, has firmly established itself as a rapid, non-destructive, and versatile cornerstone for modern food quality and authenticity testing. The integration of advanced chemometrics and machine learning has dramatically enhanced its power to deconvolute complex spectral data, enabling precise differentiation of products, detection of adulterants, and prediction of functional properties. The successful deployment of portable devices promises a future of real-time, in-field screening throughout the food supply chain. For researchers in biomedical and clinical fields, the advancements in food analysis serve as a compelling proof-of-concept. The principles and data-handling techniques explored here are directly transferable to pharmaceutical quality control, the authentication of herbal medicines, and even non-invasive clinical diagnostics, paving the way for interdisciplinary innovation. Future directions will be shaped by the deeper integration of artificial intelligence, the standardization of methods for regulatory acceptance, and the expansion of spectral libraries to create a more transparent and secure global food and health product ecosystem.

References

• 1. Spectroscopy in food and agriculture: A critical review of ... [https://www.sciencedirect.com/science/article/abs/pii/S2949824425003040]
• 2. The Role of Near-Infrared Spectroscopy in Food Quality ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11544831/]
• 3. Research progress on the application of near-infrared ... [https://bjbas.springeropen.com/articles/10.1186/s43088-025-00660-8]
• 4. Mid-Infrared Spectroscopy as a Valuable Tool to Tackle ... [https://www.mdpi.com/2304-8158/10/2/477]
• 5. Emerging roles of FTIR spectroscopy in toxic metal profiling [https://foodsafetyandrisk.biomedcentral.com/articles/10.1186/s40550-025-00119-9]
• 6. Solvent-based mid-infrared spectroscopy paired with ... [https://pubmed.ncbi.nlm.nih.gov/41185342/]
• 7. Enhancing Fourier Transform Near-infrared Spectroscopy ... [https://www.sciencedirect.com/science/article/abs/pii/S0889157525004090]
• 8. Near-infrared spectroscopy: application in ensuring food ... [https://pubs.rsc.org/en/content/articlelanding/2025/ay/d4ay02039a]
• 9. Mid-infrared metabolic imaging with vibrational probes - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC7396315/]
• 10. Regulatory Barriers: Unlocking Near-Infrared ... [https://www.spectroscopyonline.com/view/regulatory-barriers-unlocking-near-infrared-spectroscopy-s-potential-in-food-analysis]
• 11. Applications of benchtop and portable spectroscopy ... [https://www.sciencedirect.com/science/article/abs/pii/S1386142525010200]
• 12. Trends on vibrational spectroscopy tools in the agri-food ... [https://link.springer.com/article/10.1007/s00216-025-06166-7]
• 13. Infrared Spectroscopy 2025 Trends and Forecasts 2033 [https://www.archivemarketresearch.com/reports/infrared-spectroscopy-502263]
• 14. Monitoring pesticides with portable NIR spectroscopy in ... [https://www.sciencedirect.com/science/article/abs/pii/S0889157525008397]
• 15. Handheld NIR spectroscopy for real-time on-site food ... [https://pubmed.ncbi.nlm.nih.gov/40850704/]
• 16. Portable NIR Tech Detects Pesticides in Fresh Produce [https://www.spectroscopyonline.com/view/portable-nir-tech-detects-pesticides-in-fresh-produce]
• 17. Comparison of Portable and Benchtop Near-Infrared ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10024328/]
• 18. The Use of Infrared Spectroscopy for the Quantification of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10096661/]
• 19. Food Analysis with FT-NIR [https://www.bruker.com/en/products-and-solutions/infrared-and-raman/ft-nir-spectrometers/food-analysis-nir.html]
• 20. How Does NIR Spectroscopy Work For Rapid Food Quality ... [https://www.drawellanalytical.com/how-does-nir-spectroscopy-work-for-rapid-food-quality-testing/]
• 21. Use of FTIR for rapid authentication and detection of adulteration of food - PubMed [https://pubmed.ncbi.nlm.nih.gov/22129392/]
• 22. A review of Fourier Transform Infrared (FTIR) spectroscopy used in food adulteration and authenticity investigations - PubMed [https://pubmed.ncbi.nlm.nih.gov/31613710/]
• 23. An overview of recent advances and applications of FT-IR spectroscopy for quality, authenticity, and adulteration detection in edible oils - PubMed [https://pubmed.ncbi.nlm.nih.gov/33977844/]
• 24. Infrared spectroscopy: an underexploited analytical tool for ... [https://www.sciencedirect.com/science/article/abs/pii/S2214799322001552]
• 25. A Review of the Latest Spectroscopic Research in Food and Beverage Analysis | Spectroscopy Online [https://www.spectroscopyonline.com/view/latest-spectroscopic-research-in-food-and-beverage-analysis]
• 26. Using IR Spectroscopy to Ensure Food Safety [https://www.spectroscopyonline.com/view/using-ir-spectroscopy-ensure-food-safety]
• 27. Understanding near-infrared (NIR) spectroscopy in food ... [https://www.newfoodmagazine.com/article/249425/near-infrared-spectroscopy-food-analysis/]
• 28. Recent Research in Chemometrics and AI for ... [https://www.spectroscopyonline.com/view/recent-research-in-chemometrics-and-ai-for-spectroscopy-part-i-foundations-definitions-and-the-integration-of-artificial-intelligence-in-chemometric-analysis]
• 29. AI-Powered Advances in Data Handling for Enhanced ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12523821/]
• 30. Near-Infrared Spectroscopy for Honey Authentication [https://www.spectroscopyonline.com/view/near-infrared-spectroscopy-for-honey-authentication-a-practical-mini-tutorial-for-food-quality-labs]
• 31. The Application of Near-Infrared Spectroscopy Combined ... [https://www.mdpi.com/2304-8158/14/3/366]
• 32. Infrared Spectroscopy Emerges as Key Tool for Identifying ... [https://www.spectroscopyonline.com/view/infrared-spectroscopy-emerges-as-key-tool-for-identifying-plant-based-milk-alternatives]
• 33. Detection of Adulterants in Powdered Foods Using Near ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12470096/]
• 34. Decoding Plant-Based Beverages: An Integrated Study ... [https://www.mdpi.com/2673-9623/5/3/16]
• 35. Meat authenticity predictive classification by implementing ... [https://www.sciencedirect.com/science/article/abs/pii/S0889157525006775]
• 36. Spice authentication by near-infrared spectroscopy [https://www.sciencedirect.com/science/article/abs/pii/S0924224424001985]
• 37. Combating Spice Fraud Using Near-Infrared Spectroscopy [https://www.spectroscopyonline.com/view/combating-spice-fraud-using-near-infrared-spectroscopy]
• 38. In-depth chemometric strategy to detect up to four ... [https://www.sciencedirect.com/science/article/pii/S0026265X22006440]
• 39. From grain to Gain: Bridging conventional methods with ... [https://www.sciencedirect.com/science/article/abs/pii/S0956713525003512]
• 40. The Ability of Near Infrared (NIR) Spectroscopy to Predict ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8623772/]
• 41. 39 Infrared Analysis Methods [https://www.cerealsgrains.org/resources/Methods/Pages/39InfraredAnalysis.aspx]
• 42. Research Progress in the Detection of Mycotoxins ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12346599/]
• 43. Assessment of industrial cheese ripening using near ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12591493/]
• 44. Enhancing the application of near-infrared spectroscopy in ... [https://www.sciencedirect.com/science/article/abs/pii/S0308814625011057]
• 45. Strategies for analysing hyperspectral imaging data ... [https://www.sciencedirect.com/science/article/pii/S0026265X25013487]
• 46. Non-destructive hyperspectral imaging technology to assess ... [https://fppn.biomedcentral.com/articles/10.1186/s43014-024-00246-4]
• 47. Deep Learning-Enhanced Spectroscopic Technologies for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12248972/]
• 48. Advances in Hyperspectral Imaging Technology for Grain ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12428264/]
• 49. Quantification of cumin powder adulteration with rice by- ... [https://www.sciencedirect.com/science/article/abs/pii/S0889157525012876]
• 50. Non-Invasive Food Authentication Using Vibrational ... [https://www.mdpi.com/2076-3417/15/11/5948]
• 51. Application of FTIR spectroscopy in dairy products [https://www.sciencedirect.com/science/article/abs/pii/S2949824424000144]
• 52. Mini-Tutorial: Cleaning Up the Spectrum Using ... [https://www.spectroscopyonline.com/view//view/mini-tutorial-cleaning-up-the-spectrum-using-preprocessing-strategies-for-ft-ir-atr-analysis]
• 53. Preprocessing Strategies for Sparse Infrared Spectroscopy [https://pmc.ncbi.nlm.nih.gov/articles/PMC8839829/]
• 54. Mid-Infrared Spectroscopy as a Valuable Tool to Tackle Food ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7926530/]
• 55. Evaluation of a near infrared spectroscopy based method ... [https://www.nature.com/articles/s41598-025-98522-x]
• 56. Developing a near-infrared spectroscopy calibration ... [https://www.sciencedirect.com/science/article/pii/S2950289625000077]
• 57. New NIR-Based Technique Promises Rapid, Accurate ... [https://www.spectroscopyonline.com/view/new-nir-based-technique-promises-rapid-accurate-analysis-of-flavonoid-and-protein-content-in-buckwheat]
• 58. A rapid food chain approach for authenticity screening [https://pmc.ncbi.nlm.nih.gov/articles/PMC7430240/]
• 59. 2025 Review of Spectroscopic Instrumentation [https://www.spectroscopyonline.com/view/2025-review-of-spectroscopic-instrumentation]
• 60. The Role of Near-Infrared Spectroscopy in Food Quality ... [https://www.mdpi.com/2304-8158/13/21/3501]
• 61. Deep learning-based regression of food quality attributes ... [https://www.sciencedirect.com/science/article/abs/pii/S0308814625031838]
• 62. Comparative Evaluation of Near-Infrared Spectroscopy and ... [https://www.foodandnutritionjournal.org/volume13number3/comparative-evaluation-of-near-infrared-spectroscopy-and-classical-reference-methods-for-nutritional-analysis-of-fast-food-products/]
• 63. Wavelength selection enables robust quantification of oil ... [https://www.sciencedirect.com/science/article/pii/S0956713524007126]
• 64. Near-Infrared Spectroscopy Combined With An Adaptive ... [https://link.springer.com/article/10.1007/s12161-025-02901-z]
• 65. Spectroscopic techniques for authentication of animal origin ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9531581/]
• 66. Unveiling the transformative power of near-infrared ... [https://www.sciencedirect.com/science/article/pii/S0928098725001745]
• 67. 4.5 IR Data Table [https://chem.libretexts.org/Bookshelves/Organic_Chemistry/Introduction_to_Organic_Spectroscopy/04%3A_Infrared_Spectroscopy/4.05_IR_Data_Table]
• 68. Enhanced near-infrared spectroscopy for rapid and sensitive ... [https://chembioagro.springeropen.com/articles/10.1186/s40538-025-00863-2]
• 69. Comparative evaluation of various DNA extraction methods ... [https://bmcbiotechnol.biomedcentral.com/articles/10.1186/s12896-024-00933-7]
• 70. A Comparative Study on Analytical Methods by Infrared ... [https://www.ijaresm.com/a-comparative-study-on-analytical-methods-by-infrared-spectroscopy-ir-and-high-performance-liquid-chromatography-hplc-by-estimation-of-azithromycin]
• 71. Advances in Computer Vision and Spectroscopy ... [https://www.mdpi.com/2304-8158/14/3/386]
• 72. IR Spectroscopy Equipment 2025 Trends and Forecasts ... [https://www.archivemarketresearch.com/reports/ir-spectroscopy-equipment-176638]
• 73. FTIR-based molecular fingerprinting for the rapid ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12274937/]
• 74. Near-infrared Spectroscopy Market to Reach ~USD 1.3 Billion by 2035 as AI Miniaturization Expands Real-Time Quality Control [https://www.openpr.com/news/4236073/near-infrared-spectroscopy-market-to-reach-usd-1-3-billion]
• 75. Handheld NIR Technology Proves a Game-Changer in Combatting Counterfeit Cashmere [https://www.spectroscopyonline.com/view/handheld-nir-technology-proves-a-game-changer-in-combatting-counterfeit-cashmere]
• 76. The Best NIR Systems: A Buyer's Guide to Price and Features [https://www.labx.com/resources/the-best-nir-systems-a-buyers-guide-to-price-and-features/4960]
• 77. Near Infrared (NIR) Spectroscopy Equipment & Technology | VIAVI Solutions Inc. [https://www.viavisolutions.com/en-us/osp/solutions/nir-spectroscopy]
• 78. Scientists Use Water and Light to Uncover Honey ... [https://www.spectroscopyonline.com/view/scientists-use-water-and-light-to-uncover-honey-adulteration]