I. Introduction
Defect detection technology needs to determine whether there are defects on the physical object and the outer surface of the product, and identify the types of defects. When applied to products with high precision requirements, it is not only necessary to identify and classify defects, but also to accurately determine the size and location of defects. In the traditional PCB defect detection process, it is often done by comparing the tested PCB image with pre established standard templates or rules. The detection process is cumbersome[1], the rule design is complex, and the generalization is poor, which does not meet the low cost and high accuracy detection requirements in the modern industrial background; The detection method based on deep learning object detection technology can quickly and accurately detect multiple types of defects in PCB images under low-cost conditions.