Abstract
This paper presents YOLOv11-PCB, an enhanced deep learning framework specifically designed for automated defect detection in Printed Circuit Boards (PCBs). PCBs are fundamental components in modern electronics, and their reliability hinges on precise defect localization. Conventional inspection methods, such as manual inspection and traditional image processing, are limited by subjectivity, high labor intensity, and poor generalization across diverse PCB layouts. To address these challenges, we propose YOLOv11-PCB. It integrates three key innovations: (1) an Efficient Multi-Scale Attention (EMA) module for adaptive feature extraction, (2) a Content-Aware ReAssembly of Features (CARAFE) mechanism for dynamic receptive field adjustment, and (3) a refined Efficient Intersection over Union (EIoU) loss function that optimizes bounding box regression. Extensive experiments conducted on two benchmark PCB defect datasets validate the effectiveness of our proposed approach. YOLOv11-PCB achieves a mean average precision of 99.5% (mAP@0.5) and 90.7% (mAP@0.5:0.95) on the Peking University PCB dataset, reflecting a 9.7% improvement over the baseline YOLOv11. On the DeepPCB dataset, it reaches 98.9% and 81%, respectively, showing notable gains, including a 1.8% improvement over the baseline. The system maintains real-time processing capabilities at 227.2 frames per second (FPS), outperforming state-of-the-art methods in both detection accuracy and computational efficiency. These results highlight YOLOv11-PCB's robustness in identifying critical PCB defects, including solder bridges, missing components, and micro-scale fractures, while meeting the stringent throughput requirements of industrial production lines.