Abstract:China as the largest aquaculture country in the world, traditional aquaculture methods are vulnerable to light, water quality environment and complex background. In order to solve the problem of accurate feeding in fish culture and improve fish welfare, fish population was taken as the research object, and a method of fish feeding activity intensity evaluation based on feature weighted fusion was proposed by using computer vision and image processing technology. Firstly, according to the algorithm flow, the method of mean background modeling, median filtering and morphology were used to denoise and grayscale the ingested image to obtain the foreground target fish group, and the swimming trajectories of fish at different feeding stages were plotted by extracting the center of mass of the target area. Secondly, based on the pixel points of the image, the HSV color moment, canny detection and gray level cooccurrence matrix (GLCM) were used to extract the 13dimensional image features such as the color, shape and texture of the image. Then three feeding evaluation factors were selected by combining Relief feature selection and XGBoost algorithm, and the optimal weights of each evaluation factor were determined by weighted fusion method, which were 0.23, 0.40 and 0.37, respectively. Finally, the weighted fusion characteristics were compared with the traditional methods to evaluate the feeding activity intensity. The test results showed that compared with the area method, the mean square error was 0.0178, the detection accuracy was 98.89%, and the coefficient of determination was 0.9043. Compared with the traditional method based on single feature, this method not only enhanced the robustness of the algorithm, but also improved the efficiency of detection and feeding evaluation. It provided a reference for the precision feeding of aquaculture industry and the online detection of fish feeding behavior and the evaluation of feeding activity intensity.