Using the gray image of color difference G-B, Otsu algorithm was employed to segment the images of mature citrus and background. Then the occluded fruits and overlapped fruits were separated by watersheds method with distance transform. Convex hull algorithm was used to repair the occluded regions. Roundness threshold was set to remove the error segmented regions. After the citrus contours were extracted, the contours was fitted by means of the least-squares circle fit with Tukey weight function, and the centroid coordinates and radius were gained. The recognition results of 592 fruits in 87 images showed that the total correct recognition rate is up to 87.2%, and the correct recognition rate of occluded citrus and overlapped citrus are more than 80%. 