Deep Learning Based Robot Detection and Grinding System for Veneer Defects

To obtain desired wood appearance, it is necessary to grind the defects in the wood veneer. Traditional manual grinding method is time consuming and laborious. Therefore, the method using industrial robot combined with vision detection system is proposed to improve the processing efficiency. Based on the object detection network RetinaNet, the detection model is trained to detect the defects of different categories in the whole veneer. The pixel coordinates of detected defects will be transformed into robot coordinates, and PLC uses these coordinate values to control the robot for grinding. Based on the data set of veneers, experiments are conducted on the anchor boxes parameters and the weight factor of Focal Loss. The results show that the model has high recognition accuracy in the tested veneer data.