Multi-Feature Detection of Tongue Images Based on Improved YOLOv8

Tongue feature recognition, as an important component of tongue diagnosis in Traditional Chinese Medicine (TCM), has a profound and complex research background, integrating the development of traditional TCM theory with modern science and technology. This paper addresses the characteristics of tongue features and improves upon the YOLOv8 model. By incorporating the ABS module, the model's adaptability to the diversity of tongue shapes and blurred edge features is enhanced. Additionally, the original C2F structure is replaced with the C3STR module, which combines cross-stage spatial attention mechanisms with depthwise separable convolutions. This substitution optimizes feature fusion efficiency while reducing computational redundancy and enhances sensitivity to fine-grained textures. These two improvements complement each other, with the deformable convolution improving spatial modeling capabilities and the C3STR module achieving lightweight feature enhancement. Ultimately, a specialized network architecture tailored for tongue image detection tasks is constructed, providing an efficient solution for multi-target collaborative detection in complex TCM tongue diagnosis scenarios.