In order to improve locating accuracy of cutting point and avoid injury of fruit and stem for cucumber harvesting robot, an end-effector with special position feedback was developed based on machine vision. As the differences of H and S distribution on HIS space between cucumber and background, the thresholds was used for image segmentation. Subsequently, the region for robotic harvesting was extracted with edge detection on G channel of RGB space and shape features analysis. Upon the data from the firstlylocating system, the spatial coordinate of the cutting point was acquired with respect to the model of linear projection. The position correction was fed back to the robot controller. The results showed that the maximum error for the cutting point location was less than 2mm, which met the demands of the robotic harvesting operation.