Research On Multi-Label Tongue Image Recognition Based On Deep Learning

The culture of traditional Chinese medicine is vast and profound,and after thousands of years of accumulation and sedimentation,it has gradually formed a unique medical system,which has irreplaceable functions and status in treating related diseases that cannot be replaced by Western medicine.Tongue diagnosis is one of the classic methods in the four diagnostic methods of traditional Chinese medicine.Traditional Chinese medicine regards the human body as an organic whole,believing that the information of various organs converges on the tongue body through the body’s meridians.By observing the characteristic information of various parts of the tongue body,the pathological changes of each organ in the patient can be discovered.Traditional Chinese medicine tongue diagnosis is limited by doctors’ personal professional knowledge and experience,and is easily influenced by environmental and human factors,which can affect the long-term development of Chinese medicine tongue diagnosis.Therefore,the combination of traditional Chinese medicine tongue diagnosis and advanced computer technology is an inevitable trend in the modernization of traditional Chinese medicine tongue diagnosis.Traditional Chinese medicine tongue diagnosis requires analysis and research based on the multi-label category information of tongue images.However,existing tongue image research has rarely focused on the recognition of multi-label tongue images,and has neglected the correlation feature information between each label category of tongue images.To achieve high-precision recognition of traditional Chinese medicine tongue diagnosis,this paper proposes a two-stage segmentation algorithm to achieve precise tongue segmentation.Based on the characteristics of multi-label tongue image data,an improved mixed attention mechanism is used to extract the associated feature information between multi-label categories of tongue images.The main research work of this article is as follows:(1)The tongue diagnosis information is mainly concentrated in the tongue body and the edge area of the tongue body.There are many background noise factors in the original tongue image.To meet the needs of subsequent tongue image classification and recognition,the segmentation network needs to achieve accurate tongue body segmentation and edge optimization.This article proposes a two-stage tongue image segmentation network based on the characteristics of tongue image data.In the coarse segmentation stage,the tongue body is located and background noise is filtered out.In the fine segmentation stage,an improved pooling feature module is introduced to optimize the extraction of tongue edge feature information.The segmentation output is then followed by morphological algorithms to optimize detail information.(2)Tongue images are typical multi-label data,and traditional Chinese medicine tongue diagnosis requires analysis and research based on the multi-label features of tongue images.This article proposes an improved multi-label tongue image classification and recognition network model based on the characteristics of multi-label tongue image data,improves the mixed attention mechanism to extract contextual feature information of each label category in tongue images,and shares the associated feature information of each label through multi-scale feature fusion to promote multilabel tongue image classification and recognition.This article proposes a two-stage tongue image segmentation algorithm to achieve precise tongue segmentation.Through experimental comparison,it was found that the two-stage tongue image segmentation algorithm is superior to various classic deep learning network models and similar tongue image segmentation algorithm models.The algorithm results in a frequency weight intersection ratio of 95.72%,and the improved pooling module improves the overall network segmentation accuracy by0.84%.Improving the multi-label tongue image classification and recognition network model for tongue image recognition.Through comparative experiments,it was found that the F1 value of the improved multi-label tongue image classification and recognition algorithm network result accuracy is 92.47%,which is higher than the classical classification network model and the same type of tongue image classification and recognition algorithm result accuracy.The improved algorithm result accuracy is2% higher than the original basic network model.To some extent,the effectiveness and feasibility of the improved algorithm model in this paper are verified through experimental comparison.Improving the algorithm model to improve the accuracy of multi-label tongue image classification and recognition,while also laying the foundation for intelligent tongue diagnosis and computer-aided treatment,promoting the construction and improvement of traditional Chinese medicine tongue diagnosis and even the four diagnostic systems.