Research on PCB Defect Detection Algorithm Based on Improved YoloV11

Printed Circuit Board (PCB) is a core component of modern industrial electronic equipment, and its manufacturing quality directly affects the reliability and production efficiency of electronic products. Traditional PCB defect detection methods have inherent defects such as insufficient detection accuracy, poor real-time performance, and strong dependence on hardware resources. In order to overcome the limitations of current technology, based on the original YoloV11 model, this study proposes a new improved model called YOLOV11-DAM model, which introduces an improved deep separation dilated convolution (DWRSeg) module to flexibly adapt to features of different scales, thereby more accurately identifying objects in images. Secondly, ADown is used to replace the original downsampling module, which can greatly reduce the parameter value and also has a high improvement in accuracy. Finally, the Multi-Scale Dilated Transformer for Visual Recognition(MSDA) module is introduced to generate the query, key and value of the feature map through linear projection, divide the channel into multiple attention heads, and use multi-scale sliding window attention (SWDA) with different dilation rates in different heads, so as to balance computational efficiency and detection accuracy. Experimental results show that the YOLOV11-DAM model has achieved significant performance improvement on the public DsPCBSD+ dataset, with detection accuracy (Precision) reaching 85.3%, mean average precision (mAP@50) reaching 86%, and mAP (mAP50-95) covering different IoU thresholds reaching 53.4%. Compared with the original model, each indicator has increased by 1.4%, 1.2% and 1.5% respectively, making the model perform better in PCB defect detection tasks.