Research On Intelligent Tongue Diagnosis Based On Deep Learning

In recent years,Internet medical treatment has attracted widespread attention from all walks of life,and medical diagnostic methods based on artificial intelligence have gradually been recognized by the general public.As an indispensable routine examination in TCM observation,tongue diagnosis plays an important role in the clinical diagnosis of TCM.Its unique diagnosis reflects the traditional experience and characteristics of TCM diagnosis.Therefore,intelligent tongue diagnosis is also one of the hot topics in recent research.The main research content of this paper is to introduce deep learning into intelligent tongue diagnosis,segmentation and analysis of tongue images acquired under standard conditions.First,finished tongue image collection standard,collected a certain number of tongue images according to this standard,and label this batch of tongue images.Secondly,this paper proposed a tongue segmentation depth network that can effectively segment the tongue from the image.In the end,we separated of tongue texture and tongue coating based on the Lab color space and the tongue color center,and integrated deep residual network to classified the tongue color of the tongue image,and given the diagnosis result according to the corresponding relationship in the TCM theory.(1)To collate and formulate collection standards for large-scale,high-quality tongue image data sets that have not yet been published,and through cooperation with authoritative organizations,successfully constructed a number of high-quality tongue image data sets.Professional practitioners of traditional Chinese medicine label the tongue image data.In order to reduce the workload of tagging staff,an online tongue tagging system based on Bootstrap+ThinkPHP+MySQL is designed and implemented,and tongue images and their tagging results can be easily uploaded.(2)For the traditional tongue segmentation method,the segmentation accuracy is poor and it is easy to be interfered by other regions.The image segmentation based on depth model is introduced into the extraction of tongues,and the tongue segmentation network TS-Net is proposed.TS-Net can accurately segment the tongue from the original image.The experimental results show that TS-Net can achieve more than 99%of the segmentation accuracy,and the effect is much better than the existing tongue segmentation method,which is the tongue for the next step.Image analysis provides a good foundation.(3)For existing tongue image analysis methods,such as the mutual interference of tongue texture and tongue coating,or classification method is not based on the theory of TCM.This paper uses a method for separating the tongue texture and tongue coating based on Lab color space and tongue color center.Then,we combined the depth residual network to classify the tongue texture images and tongue coating images,and gave corresponding disease and diagnosis results.The experimental results show that the deep residual network separated by the fused ligaments helps to increase the recognition rate of the network and effectively improve the tongue image analysis results.