This project closely follows the hot spots of "Internet medical treatment" and the pain points of sub-healthy people, and carries out the research on intelligent physique analysis and health drink recommendation project based on multi-modal information, I .e. comprehensively diagnosing the user's symptom information and tongue diagnosis information, and recommending relevant drink information, which is presented in a small program, laying a foundation for the subsequent sale of new health drinks of "medicated diet. We cooperated with the team of Guangzhou University of Traditional Chinese Medicine to construct the tongue diagnosis data set according to the pictures collected by the cooperative hospital and the web crawler, and asked the team of Guangzhou Traditional Chinese Medicine to label them so as to improve the data set. At the same time, we also use image segmentation to reduce redundant information in tongue pictures, use data to increase and process pictures to improve robustness, and use deep learning technology for intelligent tongue diagnosis. After that, we show the early technology through small programs, and correspondingly design four functional modules: physique analysis, soup recommendation, explanation and help, and physique record. After the user enters the applet, the user can first view the instructions for using the applet. When using the constitution analysis function of the small program, you should first select the 0-4 symptoms that have recently appeared, and then under normal light, use the camera to take a complete picture of the tongue, and crop the picture and upload it. After a while, you can see the detailed constitution diagnosis results. After returning to the main interface, users can also view the results of the latest physical diagnosis and the recommended information of soup, so as to observe their recent physical changes, and find the most suitable health soup for conditioning according to their own conditions, so as to bid farewell to sub-health status as soon as possible.