This paper proposed an industrial image recognition system based on YOLOv3 in automatic production line. Complete the implementation of detection and run the identification program by using a Windows computer. The YOLOv3 is used for detection framework, and its network can directly predict the bounding box and classification probability in an image frame. Finally, the product identification results are output according to the images recognized by industrial cameras in automatic production line. Its recognition accuracy can reach 90% for magic cube.