Research Of Fine Classification Of Fat And Thin Tongue Based On Deep Convolution Neural Network

Tongue discrimination is of great significance in the diagnosis of Chinese traditional medicine.It can not only help doctors to understand the local lesions of patients,but also play an auxiliary role in the diagnosis of patients’ systemic diseases.Traditional tongue diagnosis method relies heavily on the clinical experience of the attending doctor.If the doctor’s tongue diagnosis experience is not rich enough,it is easy to make wrong judgment and thus cause misdiagnosis.With the development of computer technology,especially image processing technology and deep learning technology,the application of image processing and deep learning technology for tongue discrimination has gradually become the development trend of the industry.Based on this,this paper proposes a tongue body fat and thin fine classification method based on deep convolutional neural network,which can divide the tongue body image into four characteristics according to the tongue body characteristics: normal tongue body,fat tongue body,thin tongue body,fat tongue body and big tooth print.The main difficulty of the related work is whether it can effectively identify the feature of fat tongue and big tooth mark which is relatively fuzzy in image performance.The specific research contents are as follows:(1)Based on deep convolutional neural network,a new convolution module K Block was proposed to complete the extraction of tongue features.This module realizes feature extraction of tongue image by stacking convolution kernel of various sizes.On this basis,a deep convolutional neural network model K-CNN is built based on K Block,and classification experiments are carried out on ordinary images and Gaussian noise images respectively.The experimental results show that this model has higher classification accuracy compared with classical models such as Alex Net.(2)To further improve K Block and solve the problem that the cell1 layer of K Block is not fully utilized,a new module K2 Block is proposed.The module not only change the module within the convolution layer size,also improved the module cell1 layer and output layer of the relationship between the module and will be th e output of the cell1 layer and layer cell3 connection by the result as the output of K2 Block,the connection method of the improved residual can further improve the model fitting and gradient diffusion phenomena of the network training.On this basis,A feature extraction module and B feature extraction module were constructed based on K2 Block to extract the overall feature and detailed feature of tongue body image respectively.Finally,A new deep convolutional neural network model K2-CNN was built by combining the feature extraction results of the two.The experimental results show that the K2-CNN model combining feature extraction module A and feature extraction module B can extract more abundant and effective features,and the classification accuracy of general images and Gaussian noise images is higher than that of K-CNN model,which effectively improves the accuracy of tongue diagnosis.This paper is a new attempt of "clinical medicine + artificial intelligence",which belongs to the interdisciplinary field of medical and industrial integration,and can provide auxiliary technical support for doctors’ clinical diagnosis.