【摘要】 针对现有PCB缺陷检测算法在处理目标尺寸较小、背景线路复杂和形态不规则等场景时存在的精度不足、误检漏检等问题，本文提出一种基于改进YOLO11的PCB表面缺陷检测方法。首先，在主干网络中引入多尺度卷积注意力（MSCA）机制，增强对微小缺陷的检测能力。其次，采用高效特征金字塔网络（HS-FPN）替换传统颈部（Neck）结构，通过引入通道注意力机制（CA）和选择性特征融合机制（SFF）提升了模型对不同尺寸缺陷的检测能力。随后，结合ConvFormer与卷积门控线性单元（CGLU）改进C3K2模块，在降低计算复杂度的同时提升对有效特征的提取能力。最后，采用UIoU(Unified-IoU)损失函数动态调整候选框权重分配，提升了定位精度与收敛速度。在自建的6类PCB缺陷图像数据集上的实验表明，改进后模型的mAP达89.1%，相较原始YOLO11模型，平均检测精度提升了5.3%，精确率提升了2.3%，召回率提升了6.1%，参数量降低了28.3%，验证了所提方法在实际PCB缺陷检测中的有效性与应用潜力。更多还原

【Abstract】 To address the challenges of low accuracy, false detections, and missed detections in PCB defect inspection under scenarios with small defect sizes, complex circuit backgrounds, and irregular defect shapes, this paper proposes a surface defect detection method based on an improved YOLOv11 architecture. First, a multi-scale convolutional attention(MSCA) mechanism is integrated into the backbone to enhance the model’s sensitivity to subtle defects. Second, the neck structure is replaced with the high-level screening feature pyramid network(HS-FPN), which improves defect detection across sizes by using channel attention(CA) and selective feature fusion(SFF). Third, the original C3K2 module is enhanced by combining ConvFormer with a convolutional gated linear unit(CGLU), which improves feature representation while reducing computational cost. Furthermore, the unified-IoU(UIoU) loss function is applied to dynamically adjust the weighting of candidate boxes, enhancing both localization accuracy and convergence speed. Extensive experiments are conducted on a self-constructed PCB dataset with six defect categories. The proposed model achieves a mean average precision(mAP) of 89. 1%, surpassing the baseline YOLOv11 by 5. 3% in mAP. In addition, the precision and recall are improved by 2. 3% and 6. 1%, respectively, while the model’s parameter count is reduced by 28. 3%. The results demonstrate the effectiveness and practical potential of the proposed approach for real-world PCB defect detection tasks.更多还原