Abstract: In this paper, the apple picking robot is taken as the research object, and the apple image collected by its vision system is preprocessed and segmented. Firstly, Retinex algorithm is used to enhance the dim light, blur and other defects in the image, and bilateral filtering algorithm is used to denoise the image; Secondly, the RGB features of the color image are used to distinguish the apples and leaves in the image, and the canny edge detection algorithm is used to describe the edge information of the apples, and the minimum circle wrapping method is used to identify the radius and coordinates of the apples; Finally, in order to verify the feasibility of this method, 50 images of Red Fuji Apple collected by industrial camera were tested. The results show that the correct rate of image recognition is 86%, and the correct rate of Apple recognition is 92.8%.