RSC-UNet: An Improved UNet Method for Tongue Segmentation

Tongue segmentation is the first step in the study of tongue diagnosis and plays an important role in the study of intelligent tongue diagnosis. Due to the strong similarity in color, texture, and other information between the tongue and the colors and textures of the lips and facial regions, there are challenges in tongue segmentation. This similarity will lead to the reduction of segmentation accuracy and the loss of edge dark details. Therefore, to solve this problem, we propose a tongue segmentation method RSC-UNet based on improved UNet model. RSC-UNet employs ResNet-50 to replace regular down-sampling in UN et, which can effectively enhance the feature extraction ability and obtain better feature information. In addition, Spatial attention and channel attention are added to capture multi-stage and multi-scale tongue information on the skip connections. Experiments show that compared with DeeplabV3, UNet, PSPNet and SegFormer, RSC-UNet has better awareness of dark and edge details and presents excellent performance in tongue segmentation.