Abstract
Given the morphological similarity and medicinal efficacy differences between Acorus tatarinowii Rhizoma and Acorus calamus Rhizoma, both belonging to the Acorus rhizome slices, as well as the phenomenon of their mixed use in the market, this study aims to achieve high-precision classification and rapid object detection of these two Acorus Species Slices using deep learning technology, thus enhancing the accuracy and efficiency of Traditional Chinese Medicine (TCM) identification. The study constructed a high-quality dataset consisting of 1,928 rigorously preprocessed and annotated images of Acorus tatarinowii Rhizoma and Acorus calamus Rhizoma specimens. The ResNet50 model was employed for classification to improve classification accuracy. Furthermore, the YOLOv8 algorithm was utilized for object detection. Experimental results indicate that the ResNet50 model can accurately distinguish between Acorus tatarinowii Rhizoma and Acorus calamus Rhizoma decoction pieces, achieving a test set accuracy of 92.8%, thereby realizing precise classification. Meanwhile, the YOLOv8 algorithm achieved rapid object detection in mixed states of the two, with a detection accuracy of 98.6% and a detection frame rate of 22fps. Meanwhile, we innovatively integrate both channel attention (SE modules) and spatial attention into ResNet50 and YOLOv8 architectures, respectively, to enhance the model's ability to capture discriminative features of Acorus slices and provide a novel solution for real-time mixed-state detection.Compared to the baseline models, the SE module enhanced the classification accuracy of ResNet50 by 1.7%, while the spatial attention module improved the mAP50 of YOLOv8 by 1.2%, demonstrating the effectiveness of attention mechanisms in fine-grained identification of Chinese herbal materials.This study successfully applied deep learning technology to the classification and object detection of TCM decoction pieces, providing an effective means for intelligent identification and management of Chinese medicinal materials.