Integration of NIR spectroscopy and chemometrics for multiclass adulterant analysis in Rubus chingii Hu medicinal plant powders.

Powdered Rubus chingii Hu (RcH) is a high-value medicinal crop. Adulteration of RcH with low-value processing byproducts has become a common practice. This study combined near-infrared spectroscopy and chemometrics -partial least squares (PLS), support vector machine (SVM), random forest (RF), and backpropagation neural network (BPNN)-to establish discriminant models for RcH powder separate adulterations with powdered walnut shells (WS), tea seed shells (TS), and peanut shells (PS), as well as predictive models for adulterant concentrations. Two-dimensional correlation spectroscopy (2D-COS) analysis based on NIR and mid-infrared (MIR) spectra was used as a pre-screening tool to select the optimal spectral range required for model construction. In distinguishing specific adulterants (WS, TS, and PS) and their adulteration ratios, the SVM and BPNN models outperformed the PLS and RF models; notably, the PLS-DA model had the lowest accuracy (both training and test set accuracies were below 85%). Furthermore, this study utilized the CARS algorithm to extract feature wavelengths, thereby optimizing the model's computational efficiency and accuracy. Among the models established for predicting adulterant concentrations, the PLS, RF, and BPNN models demonstrated superior predictive performance compared to the SVM model. Considering both model performance and computational efficiency, SVM is suitable for RcH adulteration discrimination models, while PLS-R is suitable for adulteration concentration prediction. This bimodal approach provides an effective solution for combating economically driven adulteration of plant powders.