Classification Model of Ziziphi spinosae semen and Ziziphi mauritianae semen Using Hyperspectral Imaging

This study aimed to develop a hyperspectral imaging (HSI)-based classification model for distinguishing Ziziphi spinosae semen (ZSS) from Ziziphi mauritianae semen (ZMS) using AutoGluon, an automated machine learning (AutoML) framework.Conventional morphological and genetic identification methods are often costly, destructive, and time-consuming, whereas HSI enables rapid, non-contact, and nondestructive acquisition of extensive spectral information, making it suitable for large-scale verification of herbal medicine samples.A diverse set of ZSS and ZMS samples was collected, and spectral data were acquired over the 400-1,000 nm range.Sample authenticity was verified using ultra-performance liquid chromatography (UPLC) to improve dataset reliability.After spectral preprocessing using Savitzky-Golay smoothing, standard normal variate, and first-derivative transformation, AutoGluon was used to construct the classification model.The final model achieved a mean pixel-level accuracy of 98.40% ± 0.91% under 5-fold GroupKFold cross-validation on the UPLC-verified dataset.At the individual-seed level, only 3 of 169 seeds were misclassified, including 2 ZMS seeds classified as ZSS and 1 ZSS seed classified as ZMS.Permutation-based feature importance identified 628.66 nm as the most influential wavelength, and all ten most important wavelengths were located within the 596-677 nm region.Point-biserial correlation analysis showed a similar tendency in this spectral region.Misclassified seeds exhibited lower raw reflectance, occupied borderline positions between the ZSS and ZMS clusters in the preprocessed feature space, and showed lower majority-vote margins than correctly classified seeds, suggesting that the remaining errors were associated with spectrally borderline samples rather than simple brightness differences.These findings demonstrate that HSI combined with AutoGluon provides a rapid, nondestructive, and high-precision approach for authenticating ZSS and detecting ZMS adulteration, with potential applications in herbal medicine quality control and related agricultural and food sectors.