Discrimination and Quantitative Detection Models for Adulterated Dendrobium huoshanense Based on NIRS Coupled with CARS/UVE Feature Selection-Raw Data

Using 10 spectral preprocessing methods and combining two feature variable screening strategies, Competitive Adaptive Reweighted Sampling (CARS) and Uniinformative Variable Elimination (UVE), a adulteration discrimination model was established using Support Vector Machine (SVM) algorithm; Meanwhile, a pseudo quantitative model was established using Partial Least Squares Regression (PLSR).

The constructed discrimination model has an accuracy rate of 100% for identifying adulterated samples; The quantitative model has excellent performance, with coefficients of determination (R ²) above 0.98, root mean square errors (RMSE) ranging from 0.040% to 0.041%, and relative prediction difference (RPD) values of 7.699. External validation was conducted using 25 independent samples, with deviations ranging from -0.958% to 0.970%.