电瓷避雷器Issue(1)：85-94,10.DOI:10.16188/j.isa.1003-8337.2026.01.010
结合TransFormer和复合FPN的YOLOv7tiny绝缘子缺陷检测算法
YOLOv7tiny Insulator Defect Detection Algorithm Combined with TransFormer and Composite FPN
摘要
Abstract
Accurate and rapid inspection and handling of insulator defects contribute to the stable opera-tion of the power system.However,the current defect detection algorithms have problems such as com-plex network structure,long inference time,and poor robustness,which cannot meet the needs of actual inspection.Therefore,a lightweight insulator defect detection algorithm is proposed that combines Trans-Former and composite FPN.This algorithm is based on the YOLOv7 tiny framework,introducing a light-weight EMO with TransFormer architecture as the backbone network.Secondly,a composite FPN struc-ture combining context enhancement and feature refinement is designed for the neck network.Finally,Wise IOU is used as the loss function.The experiment results show that the proposed algorithm Map.5:95 achieves 70.7％,with a detection speed of 69.12frames per second,a model parameter quantity of 4.19M,and an average confidence level of 0.87 for insulator defect detection in the intuitive rendering.This indicates that the proposed algorithm achieves network lightweight,reduces the inference time,and improves detection robustness.关键词
深度学习/目标检测/YOLOv7tiny网络/绝缘子缺陷Key words
deep learning/object detection/YOLOv7tiny network/insulator defect引用本文复制引用
党宏社,许勃,张选德..结合TransFormer和复合FPN的YOLOv7tiny绝缘子缺陷检测算法[J].电瓷避雷器,2026,(1):85-94,10.基金项目
国家自然科学基金项目(编号:61871206) （编号:61871206）
陕西省科技厅自然科学基金项目(编号:2020JM-509).Project supported by National Natural Science Foundation of China(No.61871206) （编号:2020JM-509）
Natural Science Foundation Project of Shaanxi Provincial Department of Science and Technology(No.2020JM-509). （No.2020JM-509）
