Defect Detection on PCB Bare Boards Based on Improved YOLOv8 Algorithme

With the rapid development of the electronic information industry, the PCB industry, as the foundation of the electronic information industry, has a decisive impact on the quality of its products for subsequent production of electronic products. In response to the challenges of PCB defects, including small targets, diverse defect types, and unclear features, existing detection methods may face problems such as missed and false detections in actual production. In addressing this issue, this paper utilizes the PKU-Market-PCB open-source dataset from Peking University as the sample data and enhances the YOLOv8 algorithm, resulting in the YOLO-Bo network model. The algorithm introduces the BotNet in the feature extraction network, leveraging the combination of convolution and self-attention, leading to a significant improvement in the model's detection accuracy. Subsequently, considering the dataset's characteristics, a small target detection head is incorporated, reinforcing the model's capability to detect small targets. The experimental results prove that YOLO-Bo algorithm achieved a mAP value of 96.8% on the PKU-Market-PCB dataset, which is 4.6% higher than the YOLOv8n algorithm. The recall rate has improved by 6.8% compared to the YOLOv8n baseline model. This method can accurately identify defects and serve as a reference for industrial defect detection today.