An Enhanced PCB Defect Detection Approach Integrating EMA Attention Mechanism

To address challenges in printed circuit board (PCB) defect detection, such as the difficulty in identifying small targets, insufficient feature representation, and interference from complex backgrounds, this paper proposes an improved PCB defect detection method that integrates the EMA(Efficient Multi-Scale Attention) mechanism. Building upon the YOLO11(You Only Look Once) framework, this method introduces an EMA module at the higher layers of the backbone network to enhance the representation of global semantic information. Additionally, EMA attention modules are embedded into layers P3, P4, and P5 during the multi-scale feature fusion stage, thereby improving the representational capacity of features at different scales and the efficiency of information interaction. Experimental validation on public PCB defect datasets demonstrates that the proposed method outperforms baseline models in terms of precision, mAP50, and mAP50-95. Specifically, precision increases from 0.941 to 0.963, mAP50 increased from 0.956 to 0.960, and mAP50-95 increased from 0.604 to 0.610, demonstrating a significant improvement in detection performance. Although the recall rate decreased slightly, the overall detection accuracy improved significantly, indicating that this method has an advantage in reducing false positives. Furthermore, the improved model demonstrates consistent improvements across most defect categories, with particularly notable results in the detection of defects such as open circuits and pinholes. The experimental results validate the effectiveness of the EMA attention mechanism in enhancing feature representation and improving small-object detection capabilities.