Abstract: Aiming at the problem of traditional manual picking method, such as high risk coefficient, low efficiency and high labor intensity, the operation process of the picking robot was optimized and studied based on CMOS sensor. The robot was constituted of information acquisition layer, information processing layer and action execution layer. To obtain the exact location of fruits and vegetables to be picked, after image acquisition by CMOS camera, BP neural network was used to calibrate the camera, and then NDI model was used to preprocess the image to extract the useful features of the image. To verify the performance of the picking robot, fruit and vegetables identification test and distance weight measurement experiment were carried out. The test results show that the picking robot has a good recognition effect on fruits and vegetables, and the determined centroid coordinate error meets the requirements of the picking robot.