Application And Practice Of Deep Learning In Classification Of Tongue Image In Traditional Chinese Medicine

Chinese civilization has a long history.Traditional Chinese Medicine(TCM)has been spreading to the present,forming a solid theoretical basis of TCM.Different from western medicine,TCM emphasizes comprehensive understanding and systematic grasp of disease-related information,so as to guide clinical treatment,which forms a perfect complement with western medicine.The diagnostic technology of TCM is world-famous,and its unique four diagnostic methods(Inspection,Listening and smelling examination,Inquiry,Palpation)are said to be the four basic diagnostic methods of Bian Que,a famous doctor in the Warring States Period in ancient China.They are called "four diagnostic methods" according to the experience of folk circulation and his practice for many years.Tongue diagnosis is an important part of inspection,and it is also an effective means of diagnosis and treatment for doctors to understand the changes of patients' cases through tongue images.In recent years,with the deepening modernization of TCM,people have new requirements for the quantification and objectivity of tongue diagnosis.Relying on modern information technology to study the principle of tongue diagnosis,making it more scientific and convenient has become an inevitable direction of tongue diagnosis research.However,tongue diagnosis in TCM has always been judged by doctors' visual observation,and the patient's physical condition has been transformed into concise language through doctors' personal medical experience.This form of diagnosis lacks quantitative and objective measurement.At the same time,in the process of tongue diagnosis,it is vulnerable to external conditions(such as light source conditions,observation angle,etc.)and some artificial factors(such as knowledge level,clinical experience,etc.)and it often takes a long time from appointing a doctor to on-site diagnosis.However,we can also see that these unfavorable factors also create a favorable opportunity for computer-aided diagnosis.The idea of using computer to simulate the process of tongue diagnosis in TCM naturally comes into being.Taking the constitution of TCM as an example,this paper analyzed and studied the process of constitution classification based on tongue image in TCM.Through machine learning and deep learning algorithm,it realized the identification of whether there is tongue part in a picture.At the same time,it simulated the process of constitution classification of TCM by tongue image,and constructed a judgment system of constitution classification of TCM by tongue image pictures.Users can easily understand their own TCM constitution through mobile phones,and according to their own TCM constitution to obtain diet,lifestyle and other related recommendation information.The main work of this paper is as follows: 1.To collect the tongue image images uploaded from the server and the tongue image data from the outpatient department of Shanghai University of Traditional Chinese Medicine,and to calibrate the original unlabeled tongue image data according to the nine constitution classification standards of traditional Chinese medicine,a tongue image data set with constitution classification labels was constructed.2.The recognition model of tongue image is constructed.The model can accurately judge whether a picture is tongue image or not.It is realized by traditional classification method in machine learning and convolution neural network method in deep learning.The results of the two classification methods are compared and analyzed.3.Based on the structure of residual neural network in deep learning,a classification model of nine constitutions of TCM is trained.The trained model of TCM constitution classification is applied to realize an intelligent platform of TCM constitution classification.The platform can provide the classification,description and life suggestions of TCM physique for the users' tongue pictures uploaded by users.