Fusion of terahertz spectroscopy and Raman spectroscopy combined with support vector machine to distinguish different pericarpium citri reticulatae species

Pericarpium citri reticulatae (PCR), or Chenpi, is a valuable medicinal and food resource in China, but its classification can be challenging due to variations in medicinal value and market price. Traditional methods for identifying PCR involve subjective sensory analysis or cumbersome chemical techniques. This paper explores a new approach by combining terahertz and Raman spectral data with machine learning to classify seven PCR species of different ages, origins, and artificial skins. We used K-nearest neighbor (KNN), convolutional neural networks (CNN), and support vector machines (SVM) for model selection. The best model achieved 96.8 % accuracy and high AUC metrics. An innovative Manhattan data preprocessing method improved model performance by addressing errors due to instrumental and environmental factors. The T-distributed stochastic neighbor embedding (t-SNE) algorithm visualized data connections, and recursive feature elimination identified key frequencies. The THz band achieved 94.1 % accuracy using just 5.4 % of the original data, and the Raman band reached 77.8 % accuracy with 10 feature frequencies. This study demonstrates the effectiveness of spectral fusion and machine learning for precise PCR classification, offering valuable applications for traditional Chinese medicine and the food industry.