PCB defect detection based on YOLOV8 architecture

Abstract

The paper discusses the key factors and trends in the design and production of printed circuit boards (PCB), which determine the state of the art of the automatic PCB inspection. To search for and classify defects, it is proposed to use the method of detecting defects in images based on the YOLO family of object detection models. The model was trained on a public set of images of PCB with 6 classes of defects, and the accuracy was assessed using generally accepted metrics. On the test dataset, the average accuracy according to the mAP50 metric is 0.98.

Keywords

• defect
• printed circuit boards
• detection
• classification
• neural networks