Application of AFMA-Unet-Based Tongue Segmentation Model in Traditional Chinese Medicine Diagnosis

Tongue diagnosis plays a pivotal role in Traditional Chinese Medicine (TCM), yet accurate segmentation—especially of small tongues—remains a challenging task due to their diverse shapes, sizes, and feature distributions. This paper introduces a novel deep learning framework that enhances the conventional Unet architecture with an innovative Across Feature Map Attention (AFMA). By effectively capturing fine-grained features of small tongues, the proposed model significantly improves segmentation accuracy. Experimental results show that the proposed method achieves excellent performance on both tongue datasets: 97.82% Mean Intersection over Union (MIoU) on the open environment dataset which is significantly better than the existing methods. This study provides an effective solution for the automated segmentation of small-sized tongue bodies and promotes the automation of tongue diagnosis and the improvement of segmentation accuracy, which has important application prospects and positive significance for the development of the field of traditional Chinese medicine.