Qualitative identification of Bombyx batryticatus and its counterfeits using hyperspectral imaging and deep learning

Bombyx batryticatus, an important medicinal animal material in the traditional medicinal systems of China, Japan, and Korea, is in high demand due to its unique medicinal value. However, the widespread circulation of counterfeit Bombyx batryticatus (borax-treated) in the market poses a significant threat to the quality and safety of traditional Chinese medicines. This study proposes an efficient and non-destructive method for the authentication of Bombyx batryticatus by integrating hyperspectral imaging with deep learning techniques. For image-based analysis, a multi-feature fusion dataset was constructed by combining true-color composite images with features extracted from each hyperspectral channel via Principal Component Analysis. An improved ResNeXt50-KS model was proposed, integrating a Spatial and Channel Synergistic Attention (SCSA) module and Kolmogorov-Arnold network (KAN) On the fused feature dataset, the ResNeXt50-KS model achieved an authenticity classification accuracy of 98.58%. To delve deeper into the spectral characteristics of the samples, spectral data were augmented using a generative adversarial network (GAN), and a one-dimensional convolutional neural network (1D-CNN) was then constructed to automatically extract spectral features. This approach enabled both authenticity identification and geographical origin classification of Bombyx batryticatus. The 1D-CNN achieved an accuracy of 100% in authenticity discrimination. When transferred to the task of origin identification, the model maintained a high classification accuracy of 95.59%. The results demonstrate that the integration of hyperspectral and deep learning methods enables rapid and non-destructive identification of Bombyx batryticatus, providing effective technical support for quality control and market supervision of Chinese medicinal materials.