Geographical origin discrimination and quality testing of Angelica dahurica slices based on mid-infrared spectroscopy and machine learning.

Angelica dahurica, a geo-authentic medicinal herb with dual food-medicinal applications, is under key protection by the Chinese government and has gained increasing international attention. However, its quality is closely related to geographical origin, and since dried slices from different regions are visually similar, origin identification and quality control remain challenging when handling large sample sizes. This study employed mid-infrared spectroscopy combined with machine learning to discriminate the geographical origin and predict basic physicochemical indicators of Angelica dahurica slices. For origin discrimination, the Partial Least Squares Discriminant Analysis (PLS-DA) model outperformed the Discriminant Analysis Model Based on Mahalanobis Distance (DA), with the RAW-preprocessed PLS-DA model achieving a macro accuracy of 99.26% on the prediction set. The Partial Least Squares Regression (PLSR) modeling results showed excellent prediction performance for moisture with a residual predictive deviation (RPD) of 2.34, moderate performance for coumarin (RPD = 1.72), and poor performance for ash content (RPD = 1.16). Furthermore, Variable Importance in Projection (VIP) values from the optimal PLS-DA model were used to select feature wavelengths significantly contributing to classification. These were then used as input variables for both the DA model and the coumarin PLSR model. The optimal DA model with standard normal variate (SNV) preprocessing improved the overall accuracy by 16.7% to 98.0%, while the optimal coumarin PLSR model using RAW and Savitzky-Golay (S-G) preprocessing achieved an RPD of 2.22, representing an improvement of 0.50. Thus, this study proposes and validates a strategy for origin discrimination and quality testing of Angelica dahurica slices, contributing to the quality maintenance of geo-authentic medicinal materials.