Method for Classification and Origin Identification of Chinese Medicinal Herbs Based on Spectral Features

This study focuses on the discrimination of category and origin for traditional Chinese medicinal herbs using infrared spectral data. In this research, we employ advanced classification models, namely Dynamic Time Warping K-means Clustering (DTW-Kmeans), Extreme Gradient Boosting (XGBoost), and Classifier Chains, to develop models for classifying traditional Chinese medicinal herbs by both their category and origin. For unsupervised classification, we employ Piecewise Aggregation Approximation (PAA) to reduce the dimensionality of the original data and use DTW-Kmeans to quantify temporal features among different instances. For supervised classification, Principal Component Analysis (PCA) is applied to reduce the dimensionality of spectral data, and the XGBoost model is combined with five-fold cross-validation for dataset classification. For semi-supervised classification, we leverage the Classifier Chains model to handle multi-task learning and facilitate data classification. This study employs a combined approach of multiple methods for category differentiation, thereby complementing each other and mutually verifying their outcomes. Experimental results confirm the enhanced accuracy of category determination.