DBW-YOLO:改进YOLOv8n的PCB表面缺陷检测方法OA

For quality control in the manufacturing process of printed circuit board(PCB),defect detection is a critical procedure.To address the challenge of detecting surface defects on PCB and meet the demand for deployment on portable devices,the DBW-YOLO model was proposed based on the YOLOv8n model.First,a deformable attention mechanism module was introduced into the backbone network to enhance the network's focus on defect feature information by dynamically adjusting the weights of feature information.Second,in order to reduce the impact of complex environmental information,an improved modified bi-directional feature pyramid network was added to the neck network for feature fusion,which balanced the weights of different feature information.Finally,by dynamically allocating gradient increments to improve detection accuracy,the WIoU loss function was employed.Experimental results demonstrate that,compared with the YOLOv8n model,the DBW-YOLO model reduces the number of parameters and the required floating-point operations,reaching 1.8×106 and 7×109,by 40%and 13.58%respectively.It also improves the mean average precision by 2.59%,achieving 95.2%mAP.Moreover,the size of the modified model is 3.7 MB,representing only 61.67%of the original model.These results indicate that DBW-YOLO achieves high detection accuracy with low computational complexity,satisfying the practical requirements of PCB surface defect detection and making it more suitable for deployment on edge computing devices.