Classification of herbs by FT-NIR spectroscopy

Importance of the topic

Near‑infrared Fourier transform spectroscopy (FT‑NIR) provides a rapid, non‑destructive approach for identification and classification of raw botanical materials. Herbal ingredients used in nutraceuticals, cosmetics and pharmaceuticals are biologically derived and chemically complex, yet many contain diagnostically visible functional groups (hydroxyls, aromatics, amines, esters, etc.) that produce reproducible NIR signatures. Fast, robust screening of incoming herbal raw materials reduces laboratory workload, accelerates quality release and supports regulatory compliance for manufacturers that must verify botanical identity prior to extraction or formulation.

Objectives and overview of the study

This application note demonstrates that FT‑NIR combined with chemometrics can accurately classify a diverse set of 14 herbs without extensive sample preparation. The goals were to: build a calibration/classification model using raw herb samples, evaluate spectral separability between species, and validate classification performance for rapid identification of botanical starting materials used in pharmaceutical, nutraceutical and cosmetic supply chains.

Methodology and sampling

14 different herbal materials (leaves, roots, galls, peels and powdered resins) were supplied and used as received with no additional grinding or sieving. Particle size and homogeneity varied widely between classes (e.g., powdered hot pepper vs. crushed oak apple galls). Each herb class was represented by three to four samples. Spectra were acquired in diffuse reflectance mode over the near‑infrared region, and multivariate chemometric methods were applied to classify the materials.

• Spectral acquisition: Antaris II FT‑NIR Analyzer; diffuse reflectance using a closed rotating sample cup with a 30 mm window above an integrating sphere.
• Acquisition parameters: 50 co‑added scans per spectrum, 4 cm⁻¹ spectral resolution, spectral range used for modelling ~10,000–4,000 cm⁻¹ (raw spectra evaluated between 9,900 and 4,100 cm⁻¹).
• Sample presentation: closed spinning cup to improve reproducibility for heterogeneous solids; two full rotations sampled within ~1 minute.
• Software and preprocessing: Thermo Scientific RESULT Software for archiving; TQ Analyst for chemometrics. Multiplicative scatter correction (MSC, pathlength type) and a linear removed baseline were applied. No additional smoothing or derivative pretreatments were used.
• Multivariate methods: Principal component analysis (PCA) to describe variance and discriminant analysis for class assignment. Mahalanobis distance was used to quantify spectral proximity to class centers.

Used instrumentation

• Thermo Scientific Antaris II FT‑NIR Analyzer (diffuse reflectance configuration).
• Closed rotating sample cup with 30 mm window and spinner accessory; integrating sphere detector geometry for diffuse reflectance collection.
• Thermo Scientific RESULT Software for data capture and archiving; TQ Analyst for chemometric model building and classification.

Main results and discussion

The FT‑NIR method correctly classified all standard and validation spectra for the 14 herbal classes. Key analytical findings include:

• Spectral variation: Average diffuse reflectance spectra show distinctive features across herbs reflecting differences in cellulose, proteins, sugars, essential oils and other botanically relevant constituents. This variability provided the basis for chemometric discrimination.
• Dimensionality: Five principal components captured approximately 99.5% of spectral variance, indicating a compact representation of class‑relevant variation and supporting model robustness.
• Class separability: Mahalanobis distance analysis showed that for each class, the distance to the nearest incorrect class was at least twice the distance to the correct class in most cases, demonstrating clear spectral separation. All samples were correctly identified in validation testing.
• Cluster behavior: PCA score plots (first two components) revealed tight clustering for homogeneous materials (e.g., powdered hot pepper) and broader, more dispersed clusters for physically heterogeneous samples (e.g., walnut leaf, crushed oak apple galls). No class overlap was observed when the model space was examined in five dimensions.

These results indicate that despite natural heterogeneity and overlapping chemical constituents among plant materials, FT‑NIR combined with appropriate preprocessing and discriminant modeling can achieve reliable botanical classification without destructive sample preparation.

Benefits and practical applications

• Speed: Spectra collected in under a minute enable near real‑time screening of incoming raw materials at receiving docks or QC labs.
• Minimal sample preparation: Direct analysis of as‑received material avoids grinding, sieving or extraction steps, reducing labor and potential sample alteration.
• Non‑destructive: Samples are preserved for subsequent testing or reference analyses.
• Regulatory support: Rapid identity verification facilitates compliance with quality systems and pharmacopeial raw material controls.
• Operational flexibility: The approach is applicable across nutraceutical, cosmetic and pharmaceutical supply chains for botanical raw material classification and routine screening.

Future trends and opportunities

• Expanded libraries: Building larger spectral libraries covering more species, cultivars, harvest seasons and geographical origins will improve robustness and enable finer discrimination (e.g., subspecies or adulteration detection).
• Advanced preprocessing and algorithms: Incorporating advanced spectral pretreatments, variable selection and machine learning classifiers (random forests, support vector machines, neural networks) may increase sensitivity to subtle chemical differences and complex mixtures.
• Quantitation and adulteration screening: Beyond classification, FT‑NIR models can be extended to quantify key constituents (moisture, key phytochemicals) and detect common adulterants or contaminants.
• On‑line and at‑line integration: Coupling FT‑NIR with automated sampling in production lines or warehouses enables continuous quality monitoring and faster release decisions.
• Standardization and regulatory acceptance: Development of standardized workflows, validation protocols and shared spectral libraries will facilitate broader industry adoption and regulatory confidence.

Conclusion

The study demonstrates that Antaris II FT‑NIR spectroscopy, combined with MSC preprocessing and discriminant analysis, provides a fast, robust, and non‑destructive method for classifying a diverse set of botanical raw materials with complete accuracy in the tested set. The technique reduces sample handling, shortens identification time to seconds, and is suitable for routine QC and incoming material screening across nutraceutical, cosmetic and pharmaceutical contexts.

Reference

The summary is based on an application note describing classification of 14 herbs using Thermo Scientific Antaris II FT‑NIR Analyzer and chemometric analysis (TQ Analyst, RESULT Software). No external literature list was provided in the original document.