Abstract: Printed circuit boards(PCB)are widely used as core components of electronic devices. To improve the accuracy of PCB soldering defect detection algorithms,this paper proposes a YOLO-DSS soldering defect detection algorithm based on an improved YOLOv11 network. First,a dilation-wise residual(DWR)module is introduced to enhance the multi-scale feature extraction capability of the backbone network while suppressing redundant background information. Second,gated separable convolution(GSConv)is employed to construct the neck feature fusion structure,which improves multi-level feature fusion efficiency while maintaining detection accuracy. Finally,the IShapeIoU loss function is adopted to replace the conventional CIoU loss,achieving both higher localization accuracy and faster convergence. Experimental results demonstrate that,on the public PCB soldering defect dataset pcb-defect-vagef,the proposed YOLO-DSS outperforms the baseline YOLOv11,achieving a 2.1 percentage-point improvement in precision and a 2.4 percentage-point increase in mean average precision(mAP@0.5),thereby validating the effectiveness of the proposed method.