| 摘要 |
[Objectives] Issues such as adulteration and variety confusion of dried chili powder are prevalent in the current market, making the urgent development of rapid, efficient, and reliable identification methods necessary. [Methods] This study proposed a rapid method for origin tracing and variety identification of dried chili powder based on near-infrared spectroscopy (NIRS) combined with chemometric algorithms. Representative samples of four dried chili varieties (400 samples) and Erjingtiao samples from five different origins (500 samples) were collected. The samples were ground into powder, and their near-infrared spectral data were subsequently acquired. By comparing multiple preprocessing algorithms, the first-order derivative and standard normal variate (SNV) were identified as the optimal preprocessing methods for respective tasks. To eliminate the collinearity and redundant interference present in the broad absorption bands of the full spectrum, the Competitive Adaptive Reweighted Sampling (CARS) method was introduced to extract the most representative feature variables for classification. [Results] The CARS algorithm reduced the characteristic wavelengths for origin tracing and variety identification to 19 and 10, respectively, significantly decreasing model complexity. The partial least squares discriminant analysis (PLS-DA) model built on the selected wavelengths achieved perfect classification accuracy on the independent test set, with precision, recall rate, and F1 score all reaching 100%. Moreover, the extracted wavelengths were in high agreement with the absorption bands of the core components of dried chili powder from a chemical mechanism perspective. [Conclusions] This study provides a steady and lightweight theoretical basis and technical support for the real-time supervision and anti-counterfeiting tracing of dried chili powder in market sales and circulation. |
