Non-destructive identification of oolong tea varieties using near-infrared spectroscopy combined with hilbert...

To address the issue of passing off inferior products as high-quality ones in premium oolong tea markets, the study proposed a discrimination method integrating near-infrared spectroscopy (NIRS) with the hilbert bootstrapping component shrinkage algorithm (HBCSA). Given the core challenge of high spectral similarity among closely related varieties, HBCSA integrated bootstrap sampling and adaptive soft shrinkage to effectively extract deep spectral features, forming a feature set with high relevance to the target and low redundancy among features. After standard normal variate (SNV) preprocessing of the NIRS spectral data (900-1700 nm), feature dimensionality reduction was performed using CARS, BOSS, SPA, UVE, and HBCSA, followed by classification via genetic algorithm-optimized backpropagation neural network (GA-BPNN). The proposed HBCSA-GA-BPNN model demonstrated high performance, achieving a calibration set accuracy of 0.988 and optimal prediction performance with all metrics (accuracy, precision, recall, and F1-score) reaching 1.000, significantly outperforming conventional methods. These results indicate that the dual-optimization mechanism design of HBCSA offered a reliable approach for tea authentication.