Fast and accurate discrimination analysis of Angelicae Pubescentis Radix using non-targeted analytical profiles images and two-dimensional convolution neural network

This study developed an effective approach for discriminating geographical origins of Duhuo samples using non-targeted UPLC chromatograms and UV-Vis spectrogram images combined with a two-dimensional convolution neural network (2D-CNN). For comparison, four machine learning methods-extreme gradient boosting (XGBoost), random forest (RF), partial least squares discriminant analysis (PLS-DA), and support vector machine (SVM) were applied to analyze UPLC, UV-Vis data matrix, and concentrations of seven target compounds. Enhanced by data augmentation, 2D-CNN demonstrated superior accuracy, with 98.28% accuracy for UV-Vis images and 100% for UPLC images, while traditional machine learning models showed considerable variation across datasets. These results demonstrate the integration of 2D-CNN with UPLC and UV-Vis images enable robustness, accurate and non-destructive analysis for the efficient discrimination of TCM samples. Specifically, UV-Vis spectroscopy provides a convenient method for quick detection. Overall, the employed approach offers a powerful tool for the precise and reliable analysis of herbal medicines.