Tongue Image Constitution Recognition Using Improved Googlenet Model

Traditional Chinese medicine constitution is the foundation and core content of traditional Chinese medicine. It indicates the human body's susceptibility to certain specific diseases and tendency to disease types, as well as being a significant reference value for clinical medicine treatment. Tongue diagnosis is an important tool for assisting in diagnosis and recognition of constitution. Currently, constitution recognition mainly relies on doctors' professional knowledge and questionnaire methods, which takes a long time and has a low accuracy rate. In this paper, we propose a convolutional neural network model named TCR-Googlenet based on the GoogleNet model that is more accurate and effective, and the nine constitution types are recognized and classified on the constructed tongue image dataset by improving the parameters, adding batch normalization, adding attention mechanism and optimizing the structure of Inception. The validation accuracy for recognition and classification on the constructed tongue image dataset can reach 87.4%, and the validation accuracy, loss value, and convergence speed are all significantly improved when compared to the five models of VGG-16, Resnet-50, the original GoogleNet model, mobilenet-v3-large, and mobilenet-v3-small. This paper validates the proposed method experimentally, demonstrating that it is feasible and effective to use the method proposed in our study for constitution recognition by tongue images.