| In recent years,the state has made the study of TCM modernization a national strategy,and the society has paid unprecedented attention to TCM.The model of "Internet + Traditional Chinese Medicine" has great potential and broad development prospects,and will become a new development direction in the field of Internet medicine.Objective tongue diagnosis belongs to the category of intelligent diagnosis of TCM.Since traditional tongue diagnosis is always completed by doctors through visual inspection,the diagnosis results will be affected by subjective factors such as doctors’ clinical experience,visual perception and environmental factors such as light intensity,and tongue diagnosis cannot achieve objective and standardization.Tongue diagnosis in this paper,in order to realize the objective into the purpose,will be introduced to the image processing technology in the tongue diagnosis in traditional Chinese medicine,in order to improve the objective tongue diagnosis accuracy and running speed of the algorithm,based on the current tongue body and tongue segmentation algorithm as the lack of classification algorithms,key research based on the depth of the convolution tongue body and tongue segmentation algorithm of neural network classification algorithm.The main research contents are as follows:(1)To solve the problem that there is no open source data set in the field of tongue diagnosis in Traditional Chinese medicine,tongue images are collected and processed to establish tongue segmentation data set and tongue image classification data set,which are respectively used for training and verification of segmentation network model and classification network model.(2)Deep Lab V3+ was applied to tongue segmentation algorithm to improve the segmentation effect,aiming at the problem of low feature extraction ability and feature resolution caused by continuous pooled sampling based on traditional convolutional neural network.To solve the problem of large number of original Deep Lab V3+ network parameters,improve the ASPP(Atrous Spatial Pyramid Pooling)module in the encoder to reduce the number of parameters and increase the segmentation speed.To solve the problem of tongue edge segmentation,a feature fusion module was designed and added into Deep Lab V3+decoder to improve the algorithm’s ability to learn details and enrich edge feature information,thus improving the segmentation effect.The optimization algorithm Adam(Adaptive Moment Estimation)and the migration learning and training strategy are employed to speed up the convergence speed and reduce the difficulty of network training.Experimental results show that the proposed tongue segmentation algorithm based on Deep Lab V3+ is better and faster than other similar algorithms.(3)Aiming at the problem of gradient disappearance in the training process of tongue image classification algorithm based on traditional convolutional neural network,Dense Net is applied to tongue image classification algorithm to improve classification accuracy.Aiming at the problem of feature redundancy existing in Dense Net,SE module in SENet is combined with Dense Net to realize the adaptive selection of feature channels,improve the network characterization ability,and then improve the classification performance of the network.In order to further improve the classification speed of the algorithm,dense modules in Dense Net are improved and the number of model parameters is reduced by adopting deep separable convolution.The stochastic gradient descent algorithm with momentum is used to solve the problems of slow convergence speed and gradient oscillation in the training process.Experimental results show that compared with other similar algorithms,the tongue image classification algorithm based on Dense Net proposed in this paper has the characteristics of fast speed and high accuracy.Finally,a tongue diagnosis system based on Web is designed to realize the objectification of tongue diagnosis,so that users can easily obtain physical information and verify the effectiveness and practicability of tongue body segmentation algorithm and tongue image classification algorithm proposed in this paper. |
