Integrating Raman and FTIR Spectroscopy with Machine Learning for Component Profiling and Geographical Origin Identification in Astragalus

Astragalus is an economically significant medicinal plant widely utilized in traditional medicine due to its remarkable immunomodulatory, anticancer, and antiviral properties. In this study, we propose an innovative approach that combines spectroscopic techniques (Raman spectroscopy and Fourier Transform Infrared Spectroscopy, FTIR) with Principal Component Analysis and Support Vector Machine (PCA-SVM), and Convolutional Neural Networks (CNNs) for the precise classification and identification of different parts of Astragalus from various regions. The classification accuracy using the PCA-SVM algorithm exceeds 90%, while the CNN algorithm achieves classification accuracy over 99%. Additionally, through the internal standard method, spectral features reveal compositional differences of cellulose, lignin, astragaloside, isoastragaloside, and flavonoids across the xylem, phloem, and periderm of Astragalus. Therefore, this non-invasive, rapid, and cost-effective detection method is expected to be applied to the study of Astragalus' multilayer structure and the identification of its geographical origin.