CDA-YOLO: an improved YOLOv11 model for PCB defect detection

Printed circuit boards (PCBs) are fundamental to modern electronic systems, and their manufacturing quality directly determines the performance and reliability of the final product. However, existing YOLO‑based defect detection models still face critical challenges, including high miss rates for tiny defects and frequent false positives caused by complex background textures. To address these issues, we propose CDA‑YOLO, an enhanced YOLOv11 framework incorporating three key innovations. First, a Cross-level Bidirectional Feature Pyramid Network (CBiFPN) is designed, which effectively alleviates the loss of fine-grained details of small targets as features propagate through deep layers. Second, a lightweight Dual-Scale Convolution module (DualScaleConv) is proposed, which employs a parallel branch structure to simultaneously capture local details and global contextual information. Finally, a parameter-free average attention mechanism (PfAAM) is integrated into the C3k2 module to enhance the feature response in defect-related regions and suppresses background noise. Experimental results demonstrate that the proposed model achieves competitive performance on the PCB defect dataset, with mAP@0.5 reaching 96.1% and mAP@0.5:0.95 reaching 56.0%, corresponding to improvements of 2.8% and 3.4% over the baseline YOLOv11, respectively. These results validate that CDA-YOLO offers an efficient and robust solution for industrial-grade PCB defect detection.