Research on PCB Small Target Defect Detection Based on Improved YOLOv5

Abstract

As global automation accelerates, the importance of the PCB as a core component of electronic products grows with each passing day. The smallest hazards in PCBs can cause huge losses, so testing the quality of PCBs is an important step in the production process. To address the high level of integration, miniaturization, and multilayering of PCB production technology, we are using a new and improved model based on YOLOv5 to detect PCB defects. This new model solves the problems of difficult feature extraction, the similarity between features, and poor detection performance of PCB defects. In this paper, we use 10,668 images of PCB data containing six different defects. Experimental results show that the improved model in this paper has a detection accuracy of 99.0% and a detection speed of 0.016s compared to other defect detection algorithms of the same type.