PEAN-YOLO: Improved YOLO11 for Small-Scale PCB Defect Detection

PCBs serve a critical role in modern electronic devices. Reliable defect detection before entering the market is essential. However, PCB defects are often small, making accurate detection a major challenge. To address this issue, based on YOLO11n, we propose an efficient network named PEAN-YOLO. The proposed model introduces four core optimizations: upgrading the C3k2 module with PConv for enhanced feature extraction, incorporating an additional high-resolution branch (P2, 160×160) into the multi-scale feature fusion network to better preserve fine-grained spatial details, inserting novel ACPCA modules before each detection head to enhance focus on defect regions, and integrating CIoU with NWD into the bounding box regression loss for more precise localization. Evaluated on the augmented PKU-Market-PCB dataset, PEAN-YOLO achieves a competitive mAP50 of 96.6% with low Parameters and FLOPs, demonstrating superior detection performance and excellent computational advantage.