Automatic Recognition of Liver Depression Line in the Tongue Based on the Self-Attentive Mechanism of Vision Transformer
Automatic recognition of tongue images is the core issue of the intelligent tongue diagnosis,which is a typical application of computer vision in the field of tra-ditional Chinese medicine.The liver depression line(LDL),which is located on both sides of the tongue surface,is an important clinical diagnostic basis of traditional Chi-nese medicine.Its automatic visual recognition can promote the further development of intelligent tongue diagnosis.In order to improve the accuracy of LDL recognition,we propose an attention-guided and local region reuse strategy based on the advan-tages of the self-attention mechanism of the visual Transformer.We then design a network,ST-LDL,for LDL recognition based on Swin Transformer.The attention guidance strategy is used to promote ST-LDL to extract fine-grained features of LDL.The local region reuse strategy utilizes the local strong response areas on both sides of the tongue surface to train the local network branch of ST-LDL on one hand and enhance the global network branch of ST-LDL on the other hand.Ablation exper-iments show that using attention guidance and policy region reuse strategies alone can improve recognition performance;all metrics increase significantly when these two strategies are shared.Comparison experimental results show that our algorithm outperforms other existing advanced algorithms.
Intelligent tongue diagnosisliver depression linevision transformerself-attention mechanismattention guidancelocal area reuse
NETL
NSTL
万方数据
