Abstract: Citrus, as an important fruit in our country, plays an important role in increasing the income of our fruit farmers. Therefore, automatic robots of citrus picking have become a hot topic in recent research. A citrus image recognition algorithm was designed for the core machine vision part of the citrus picking robot. The S-component value in HSV color space was selected as the color feature for citrus recognition to perform image enhancement, denoising and segmentation pre-processing operations. By comprehensively comparing the Hough transform under the three edge operators Canny, Sobel and LOG, we finally determined the optimal solution for citrus image recognition. The experimental results show that the Hough transform based on the Sobel edge detection operator is the best, with a success rate of 97% for a single citrus without branch and leaf shading, 90% for a single citrus with branch and leaf shading, and 80.6% for multiple overlapping citrus.This algorithm can effectively guarantee the probability of successful identification of citrus to a certain extent.