智能系统学报2025，Vol.20Issue(5)：1082-1092,11.DOI:10.11992/tis.202406045
基于阶梯式特征融合的输电线路外力破坏检测
Detection of external force damage of transmission lines based on stepwise feature fusion
摘要
Abstract
A detection method based on stepwise feature fusion is proposed to address the challenges posed by instru-ment deformation under unmanned aerial vehicle(UAV)shooting angles and complex features caused by different oper-ating states of arm-bearing machinery during transmission line inspections.The method integrates deformable large ker-nel attention networks to extract features from UAV images,performs stepwise feature fusion using a multiscale se-quence feature fusion module,and applies lightweight operations to the detection head to reduce the parameters.An gradual complete intersection over union non-max-imum suppression(GCIoU NMS)loss function further enhances the model.Experiments on a self-constructed dataset show mAP50％and mAP50％-95％improvements of 10.5 and 10.2 points,reaching 86.8％and 58.4％,respectively.On the VOC dataset,mAP50％and mAP50％-95％increased by 7.3 and 8.1 points to 79.5％and 58.8％,respectively.The results demonstrate the effectiveness of the method for external force damage detection of transmission lines in complex environments.关键词
输电线路外力破坏/目标检测/特征提取/阶梯式特征融合/检测头轻量化/GCIoU损失函数Key words
external force damage of transmission lines/object detection/feature extraction/stepwise feature fusion/lightweight head/GCIoU loss function分类
信息技术与安全科学引用本文复制引用
赵文清,蔡建颖,李赛辰..基于阶梯式特征融合的输电线路外力破坏检测[J].智能系统学报,2025,20(5):1082-1092,11.基金项目
国家自然科学基金项目(62371188) （62371188）
河北省自然科学基金项目(F2021502013). （F2021502013）
