Discrimination of Panax ginseng powders by botanical part using multi-spectroscopic data fusion and a satisfaction grade strategy

Different botanical parts of Panax ginseng (main root, fibrous root, and rhizome) exhibit varying active component profiles and commercial values. Once processed into powders for functional foods, they become visually indistinguishable, making them highly susceptible to economically driven adulteration. To address this, a rapid identification methodology was developed utilizing near-infrared (NIR), mid-infrared (FTIR), and Raman spectroscopy based on a robust dataset of 386 authentic sample batches. To explore and establish the optimal single-spectrum models, a primary cohort of 326 batches was utilized. Through the systematic optimization of spectral preprocessing and feature extraction (CARS, PLS), highly robust base classifiers (SVM, GWO-SVM, LSSVM) were successfully constructed. To overcome the inherent analytical bottlenecks of single-modality analysis, an efficient decision-level fusion architecture adapted for multi-spectroscopic chemometrics, based on a “Satisfaction Grade”, was constructed. This non-parametric mechanism dynamically standardizes the geometric distances from samples to separating hyperplanes into soft-voting weights. During external testing, this confidence-weighted framework neutralized individual spectral errors, achieving a flawless 100% classification accuracy on an independent validation set of 60 samples. By reducing the risk of fusion-layer overfitting, this coupling strategy demonstrates significant application potential for the authentication of Panax ginseng powders from different botanical parts and their related quality control.