Classification of herbs by FT-NIR spectroscopy

Classification of herbs by FT-NIR spectroscopy — concise expert summary

Significance of the topic

FT‑NIR spectroscopy provides a fast, non‑destructive approach for raw material identification based on molecular vibrational signatures. For industries using botanicals (pharmaceutical, nutraceutical, cosmetic), reliable and rapid identification of incoming herbs reduces reliance on slow, labor‑intensive techniques (morphology, TLC, HPLC), supports regulatory compliance, and improves supply‑chain quality control. This work demonstrates that FT‑NIR combined with multivariate classification can distinguish complex herbal matrices despite shared biochemical components such as cellulose, proteins and sugars.

Objectives and study overview

• Demonstrate rapid classification of a representative set of commercial herbs using FT‑NIR spectroscopy and chemometrics.
• Assess robustness of a discriminant model for heterogeneous botanical raw materials without extensive sample preparation.
• Quantify class separation and validation performance using principal component analysis and Mahalanobis distance metrics.

Methodology

• Samples: Fourteen different herbs and plant parts (examples: agrimony, chamomile flowers, gentian root, hot pepper powder, walnut leaf, oak apple gall, myrrh) were provided by IREL and used as received. Particle size varied widely (thin leaves up to ~6×6 mm; crushed roots/galls up to ~10×10×5 mm). No additional grinding or sieving was performed.
• Spectral acquisition: Each sample was measured in a closed rotating sample cup positioned over an integrating sphere. Spectra were acquired from approximately 10,000–4,000 cm⁻¹ (calibration used 9,900–4,100 cm⁻¹), 50 co‑added scans per spectrum at 4 cm⁻¹ resolution. The cup spinner enabled two full rotations and acquisition in under one minute per sample.
• Preprocessing and chemometrics: Multiplicative signal correction (MSC) of pathlength effects and a linear baseline removal were applied. No smoothing or derivative preprocessing was used. Discriminant analysis (via Thermo Scientific TQ Analyst) built classification models. Principal component analysis (PCA) summarized variance; five principal components explained ~99.5% of spectral variability.

Used instrumentation

• Thermo Scientific Antaris II FT‑NIR Analyzer with integrating sphere and closed rotating sample cup (spinner).
• Thermo Scientific TQ Analyst software for chemometric model development (discriminant analysis) and RESULT software for data collection and archiving.

Main results and discussion

• Classification performance: All standards and validation spectra were correctly identified by the discriminant model. Validation used randomly chosen spectra from each class.
• Spectral variability and PCA: Representative diffuse reflectance spectra showed exploitable differences among the 14 herbs despite shared biochemical constituents. PCA indicated that five components captured ~99.5% of variance, and two‑dimensional score plots revealed well separated clusters for most classes.
• Cluster behavior: Homogeneous, finely powdered materials (e.g., hot pepper powder) produced tight clusters, while heterogeneous or irregularly shaped materials (e.g., walnut leaf, crushed oak apple gall) produced more dispersed clusters. Nevertheless, clusters remained separable in multidimensional principal component space.
• Mahalanobis distance analysis: For each class the Mahalanobis distance to the nearest incorrect class was generally at least twice the distance to the correct class, indicating good class separation and robustness to spectral variability. Tabulated distances quantified how close samples would be to incorrect classes; low within‑class distances and substantially larger nearest‑incorrect distances supported reliable classification.

Benefits and practical applications

• Speed and throughput: Spectra acquired in under one minute per sample enable high throughput incoming raw material screening.
• Minimal sample preparation: The method worked on materials as received, avoiding time‑consuming grinding or sieving steps and reducing risk of sample alteration.
• Reproducibility for heterogeneous samples: The closed rotating cup with spinner produced reproducible spectra from nonuniform botanical materials.
• Operational value: The approach is suitable for routine QA/QC checks, raw material identity testing, and rapid triage prior to more detailed laboratory analyses.

Future trends and potential applications

• Broader calibration libraries: Expanding the spectral library to cover more species, geographic/seasonal variability, and adulterants will improve robustness in industrial settings.
• Advanced chemometrics and machine learning: Implementation of supervised machine learning, ensemble methods or deep learning could improve classification accuracy and handle greater sample heterogeneity.
• Instrument transfer and standardization: Developing transfer protocols and spectral standardization will increase applicability across instruments and sites.
• Integration into PAT and supply‑chain systems: Inline or at‑line NIR solutions and database linking can enable real‑time raw material control and traceability.
• Data fusion: Combining FT‑NIR with orthogonal techniques (e.g., chromatography, mass spectrometry) can enhance authentication and detect subtle adulteration.

Conclusion

FT‑NIR spectroscopy combined with discriminant analysis provided a rapid, reproducible, and accurate method to classify a diverse set of 14 herbs used in nutraceutical, cosmetic, and pharmaceutical raw materials. The method required minimal preparation, achieved complete correct identification in validation, and showed clear spectral separation between classes in multidimensional space. While heterogeneous botanical materials exhibit greater spectral scatter, the sampling accessory and chemometric approach delivered robust classification suitable for routine QA/QC applications.

References

• Hollein M., Strother T., Classification of herbs by FT‑NIR spectroscopy, Thermo Fisher Scientific Application Note AN51724_E (2022). Contributions from IREL, spol. s.r.o. (Brno, Czech Republic).