Current methods for detecting weld defects on the surfaces of petrochemical storage tanks suffer from low efficiency and are prone to missed and false detections,especially for small targets.To address these problems,this paper proposes YOLOv8-SDB,an improved YOLOv8n-based method for weld defect detection.First,a symmetric positive definite convolution (SPDConv) module is incorporated into the backbone network to mitigate the loss of detail information during shallow feature extraction,thereby capturing richer spatial and channel information.Second,a weighted bidirectional feature pyramid network (BiFPN) is utilized to fuse multi-level features,enhancing feature representation capability.Third,the lightweight and efficient DySample (dynamic sampling) upsampling operator is adopted to improve feature reconstruction capability while reducing computational complexity.Finally,the weighted intersection over union (WIoU) loss function is employed to accelerate the convergence of bounding box regression loss and improve localization accuracy.Experiments conducted on a weld defect dataset demonstrate that the proposed YOLOv8-SDB algorithm achieves a precision of 86.2%,a recall rate of 79.4%,and a mean average precision (mAP) of 84%.These results represent improvements of 3.4%,2.8% and 3.9%,respectively,over the baseline YOLOv8n algorithm.