Non-destructive classification of Peruvian medicinal plants using NIR hyperspectral imaging (1300–1700 nm) and machine learning

Introduction. The identification of native medicinal plants often relies on destructive, time-consuming, or expert-dependent methods. This study proposes the use of near-infrared hyperspectral imaging (NIR-HSI) and machine learning as alternative, non-invasive tools for the rapid discrimination of three medicinal species from northern Peru: Peperomia inaequalifolia (congona), Alternanthera sp. (lancetilla), and Teline monspessulana (retama). Objectives. To evaluate the performance of multiclass classification models applied to preprocessed NIR-HSI spectra, aiming to develop a reliable system for plant identification and quality control. Materials and Methods. A total of 1467 spectra were collected using a NIR-HSI camera in the range of 1300–1700 nm. Spectral data were preprocessed using Savitzky–Golay smoothing and standard normal variate (SNV). Seven machine learning classifiers were trained and evaluated through stratified 5-fold cross-validation, including Random Forest, Gradient Boosting, XGBoost, and Ridge Classifier. Results. Random Forest achieved the highest performance (accuracy = 0.9959), with a ROC-AUC of 1.00. The remaining models yielded mean accuracies ranging from 0.9720 to 0.9945, with ROC-AUC values close to 1.00, indicating strong discriminative capability. Conclusions. The combination of NIR-HSI and supervised learning models enables highly accurate classification of medicinal plants. This approach shows potential for traceability, quality assurance, and ethnobotanical validation, particularly in decentralized or resource-limited settings.