The constitution is formed by the congenital inheritance and acquired. The individual constitution is different, and the physiological response to the outside world is different. Therefore, research on physical classification can better prevent diseases. In order to improve the accuracy of physical classification, while overcoming defects such as slow convergence speed in the neural network, a gray wolf algorithm (GWO) optimized convolutional neural network (CNN) and support vector machine (SVM) human constitution classification method. Most researchers have collected physiological signals as assessment parameters, so we use ECG as a classification, combining tongue image diagrams as a basis basis, and more comprehensive classification of human constitution. First of all, the experimental data is extracted, and secondly, the GWO optimized CNN-SVM is classified and identified by the data sample. The final experimental results are compared with other classifier models. The classification of the classifier designed in the text is accurate.