Research on PCB Surface Defect Detection Based on Improved YOLOV8

With the rapid advancement of industrial automation technology, the quality of PCBs directly affects the performance of electronic products. To solve the problem of missed or repeated detections caused by unreasonable anchor frame design in PCB surface defect detection tasks based on anchor frame detection algorithms, this paper proposes a BBW-YOLOv8s based anchor frame free PCB surface defect detection model using the one-stage anchor frame free algorithm YOLOv8s as the benchmark network. The model first adds a Biformer attention mechanism with good performance for small object detection to the SPFF module and Neck layer of the YOLOv8s network, reducing the model's attention to irrelevant information in PCB images and enhancing its ability to learn key features; Secondly, referring to the multi-scale feature fusion network Bi FPN, the output of shallow feature maps in the backbone network is connected across scales with the detection head, so that the detailed information of shallow feature maps is guided by the semantic information of deep feature maps, and the model's understanding ability of PCB image content is improved; Finally, the cross entropy loss function in the original YOLOV8 model was replaced with Focal Loss to make the model more focused on difficult to classify targets, thereby improving overall accuracy.