Abstract: In order to improve the intelligent manufacturing level of steel coil production,the galvanized sheet production line of a steel plant in Guangzhou add a sampling robot based on visual guidance for automatic labeling and handling of steel plates.The vision system based on lines_gauss function is developed for edge detection of steel plate.Although it can satisfy the calculation of center coordinates of steel plate in most scenarios,the recognition accuracy rate is still not above 99%.The ResNet-18 deep convolutional neural network and Faster R-CNN object detection model are used to enhance the detection ability of the visual system to the edge of the steel plate,which can solve the problems caused by uneven illumination,large differences in the reflectance of different coated steel plates,and difficulty in detecting the edge of stacked steel plates.