| Artificial intelligence technology is widely used in various industries,and Artificial intelligence medical treatment gradually attracted widespread attention.With the aid of artificial intelligence,the preliminary screening can be completed quickly,the misdiagnosis rate can be reduced,and the speed of doctors’ reading medical images can be improved.Tongue diagnosis,as an important part of traditional Chinese medicine,is simple and easy,and can quickly understand the disease.Intelligent tongue diagnosis is also one of the hot research contents.With the maturity and development of deep learning theory and model,its application in tongue segmentation also has a good effect.However,when the data comes from different data distribution,the performance of the model is seriously degraded.To solve this problem,this paper proposes a cross-domain adaptive tongue segmentation framework,which can still have a good tongue extraction effect in the tongue image data from different domains.Besides,there is little research on the tongue with teeth printed and cracks,which is not conducive to the objective development of tongue diagnosis in traditional Chinese medicine.Therefore,this paper proposes a weakly supervised machine learning algorithm to detect the tongue with teeth printed and cracks in the tongue image,which can greatly improve the detection model in the environment of only a few labeled tongue image data The detection effect of tongue image.In this paper,the segmentation of the tongue image and the detection of the crack and the tooth trace of the tongue image are studied deeply,and the intelligent tongue image analysis system is designed and implemented according to the research content.The main research work includes three parts:(1)the domain adaptive problem of tongue image segmentation;(2)the problem of tongue image crack detection based on weak supervision;(3)the realization of intelligent tongue image analysis system.The main innovations of this paper are as follows:(1)To solve the domain adaptive problem of tongue segmentation,an iterative cross domain tongue segmentation framework is proposed.Inspired by the snake method,this paper proposes a tongue image evaluation method by analyzing the characteristics of the tongue and combining the contour of segmentation results.Without the ground-truth label,it can evaluate the result automatically.Using the tongue image evaluation method to select better prediction results as pseudo tags,and integrating the original data set adjustment model,the generalization ability of the deep learning model on new data is improved.The experimental results show that the framework can deal with the cross domain adaptive problem of tongue segmentation well,and the dice score of the model is improved from 70.11%to 98.26%,which is better than the existing methods.(2)To solve the problem of the large workload of labeling data,a weak supervision detection method is proposed.Only a small amount of data needs to be labeled completely,and the rest data need to be labeled with image category information.While learning the data with complete annotation information,this method uses image category information to learn the features without detailed annotation data,improves the detection ability of the model,and alleviates the overfitting problem caused by too little complete annotation information data.The experimental results show that this method can reduce the workload of marking,and at the same time,it can detect the tooth trace and crack in the tongue image.Combined with the existing tongue image data,the existing tongue image segmentation and the detection of teeth printed tongue and fissured tongue methods,an intelligent tongue image analysis system based on Django+Ajax+vue.js+MySQL+Python is designed and implemented.Through this system,the user can upload the tongue image,segment the tongue image or detect the tooth mark crack according to the needs,and observe the change of the condition by viewing the historical records. |
PDF Full Text Request