Humans have the ability to quickly identify their own environment, understand, judge and analyze it, which is one of the important reasons why human beings can survive in nature for a long time and gradually develop it into a prosperous society today. The key to human's ability to perceive and understand the environment lies in the ability to accurately nd and identify objects, understand and describe visual scenes, and even express emotions on this basis. And if computers can realize automatic and accurate image recognition, and even understand the semantics of images correctly, it will surely improve and facilitate human life. One of the research hotspots in this eld is the human visual system, so one of the important research directions of new science and technology is image recognition. In this context, this paper completes the design and improvement of the computer image recognition system. The core technology of the system is to improve the image algorithm, which can complete the training, testing and classi cation of target images. The experimental data is available.Choosing this algorithm to improve the learning and training of the data generated by the original image processing is more effective than directly training the original image.