Classification Of Tongue Pattern Based On Convolution Neural Network

Tongue diagnosis is one of the special diagnosis methods of traditional Chinese medicine.It is widely used in the clinical research and practice of traditional Chinese medicine.Tongue color,tongue coating color and tongue coating quality are important features of tongue.The extraction effect of these features will affect the accuracy of final tongue diagnosis.In this paper,the convolution neural network,which is rising in recent years,is applied to the classification of tongue pattern,and a classification model of tongue pattern based on convolution neural network is constructed.Different from the previous classification of tongue pattern,which is necessary to extract the feature of tongue image artificially by human vision,the convolution neural network can automatically extract the features of the tongue image by a series of hierarchical abstract processing to the input tongue image,and then directly output the tongue pattern at the output end.By this way,the automatic classification of tongue pattern is realized,and the clinical diagnosis efficiency and ability of tongue diagnosis can be improved to a certain extent.In the process of diagnosis,530 tongue images are collected in real time,the diagnosis results such as tongue color,tongue coating color and tongue coating quality are recorded,and the data set of tongue image is set up.After a series of preprocessing,the data is applied to training and validation of tongue image pattern classification model.Then a new classification method of tongue pattern based on single task convolution neural network is proposed.Five convolution neural network models are constructed for the classification of tongue color,tongue coating color,tongue coating thickness,tongue fur moistening and tongue fur grease.These models use the color tongue image as input,transform the features of tongue by middle layers and output the corresponding tongue pattern.The classification accuracy of the five models on the test set is more than 80%.Compared with other common classifiers,this method has a considerable advantage in accuracy.Then a classification method of tongue image pattern based on multitask convolutionneural network is proposed,and five classification tasks are completed by two multi task convolution neural network models.Two color classification tasks of tongue color and tongue coating color are correspond to one network model,and three texture classification tasks of tongue fur thick,tongue fur moistening and tongue fur grease are corresponded to another network model.Each network model is trained by multiple tongue pattern classification tasks at the same time.The trained model has the ability to complete multiple classification tasks at the same time.Compared with single task convolutional neural network,this method has the advantages of saving time and saving computation,without sacrificing the accuracy of classification.Finally,a graphical user interface is designed to achieve the choice of tongue image and the display of classification results.