Research on Chinese herbal medicine image detection based on improved YOLOv8 network

We present an advanced image recognition algorithm for Traditional Chinese Medicine (TCM) materials based on an improved YOLOv8 architecture. The proposed algorithm addresses the limitations of traditional models in TCM image recognition by integrating a CSRestormer module and an F-CMRF module, which enhance the representation of medicinal features. This design enables the model to simultaneously capture global contextual information and fine-grained structural details. Evaluations on the TCM object detection dataset cm_detect44 demonstrate that the algorithm achieves an mAP50 of 89.2%, surpassing the baseline YOLOv8n model's 88.3%. These results validate the enhanced performance of our method in TCM image recognition tasks.