Application of a semi-supervised learning framework combining principal component constraint enhancement in near-infrared spectroscopy of Rhizome Coptidis.

Rhizome Coptidis is a popular traditional Chinese medicinal herb, and analyzing its composition is crucial for evaluating its quality. However, existing analytical methods can produce significantly different results due to various practical factors, so there is an urgent need to develop more suitable analytical methods. Near-infrared spectroscopy (NIR) is widely used in the analysis of traditional Chinese medicine due to its rapid, non-destructive, and pollution-free characteristics. It offers numerous possibilities for the detection of Rhizome Coptidis. However, this technology often encounters challenges such as small sample sizes, limited annotated data, and redundant information. More advanced modeling methods are therefore required to address these issues. Based on this, this paper proposes a fusion framework for the quantitative analysis of Rhizome Coptidis: Spectral Augmented Principal Component Analysis-Semi-Supervised Learning-Convolutional Neural Network-Twin-Kernel Support Vector Regression (SAPCA-SSLCNN-TKSVR). This framework comprises three modules: Spectral Augmented Principal Component Analysis (SAPCA) for data augmentation, Semi-Supervised Learning Convolutional Neural Networks (SSLCNN) for feature extraction, and Twin Kernel Support Vector Regression (TKSVR) for data prediction. The experimental results demonstrate that this method is well-suited to the quantitative analysis of Rhizome Coptidis, particularly in scenarios involving small sample sizes, limited labeled data, and redundant information. It effectively improves modeling performance for high-dimensional, highly linear data and offers a viable approach to the field of near-infrared spectroscopy.