基于改进YOLO v8n的聚乙烯管道焊缝检测算法研究与应用OA

Detecting weld defects in polyethylene pipelines remains challenging due to low defect visibility,complex back-ground interference,and high variability in defect morphology as factors that traditionally necessitate extensive human ex-pertise.To address these limitations,this study presents FWD-YOLO,an improved object detection model based on YOLOv8n,specifically tailored for weld defect inspection.Key enhancements include:(1)replacing the standard Conv and C2f modules in the backbone with more efficient alternatives;(2)integrating the dynamic sampling module DySam-ple into the neck to improve robustness against geometric deformations and cluttered backgrounds;(3)introducing a multi-layer channel attention mechanism to facilitate effective multi-scale feature fusion;and(4)optimizing the loss func-tion to better balance precision and recall.Evaluated on a self-constructed dataset encompassing nine distinct defect cate-gories,FWD-YOLO achieves a 4.8%absolute improvement in mAP@50(reaching 73.3%),a 3%gain in recall,while simultaneously reducing model parameters by 22.4%and floating-point operations(FLOPs)by 42.6%com-pared to the original YOLOv8n.Furthermore,FWD-YOLO outperforms other lightweight YOLO variants,including YOLOv5n,YOLOv6n,YOLOv9t,YOLOv10n,YOLOv11n,and YOLOv12n,in terms of detection accuracy.These results demonstrate that FWD-YOLO offers an efficient,accurate,and deployable solution for automated quality inspection of polyethylene pipeline welds,thereby contributing to the safe and reliable operation of pipeline infrastructure.