Abstract: Printed Circuit Boards (PCBs) are crucial electronic components with a significant impact on the quality of electronic products. However, existing PCB defect detection algorithms suffer from limitations in detection accuracy, especially in precise defect localization. To address these challenges, this paper proposes a novel PCB defect detection algorithm based on EfficientNetV2. Building upon the Faster R-CNN architecture, we leverage the powerful feature extraction capabilities of EfficientNetV2_M as the backbone network. Furthermore, we optimize the feature fusion network FPN (feature pyramid network) by incorporating the Channel Attention Mechanism (ECA) to enhance the extraction of intricate details. Experimental results on the PCB defect data set released by Peking University Intelligent Robot Open Laboratory demonstrate that the proposed defect detection algorithm outperforms the state-of-the-art LWN-Net in terms of mAP (mean Average Precision) at IoU=0.50, with an increase from 99.58% to 99.66%. Moreover, at IoU=0.50:0.95, the mAP significantly improves from 52.6% to 79.4%. Consequently, the enhanced algorithm achieves highly accurate PCB defect detection while addressing the issue of imprecise defect localization, thus holding substantial practical significance. The code is available at https://github.com/ChaO989/Defect_detection.