Weak Weld-target Recognition Based on Prior Knowledge

The welding technology is widely used in many manufacturing industries. The welding seams are weak targets due to unstable welding quality and the various kinds of noises. This article proposed a new method of weak-target recognition focus on welding seams, which helps to automatically track the welding seam in complex enviroment. We first relabel the data with “outer boxes” and “slopes” on the welding seam. Then we design a new recognition framework based on prior knowledge and YOLOv5 recognition algorithm. We trained weld images in many kinds of enviroment and compared the results. It shown that our method exceed the tradional method on both precision and recall.