基于改进YOLOv7-tiny的绝缘子缺陷检测网络

打开文本图片集

中图分类号：TN919.85-34；TP391.41 文献标识码：A 文章编号：1004-373X（2025）16-0105-08

Insulator defect detection network based on improved YOLOv7-tiny

HANXingyu，CHENWeizhen （SchoolofElectricalandElectronicEngineering，WuhanPolytechnic University，Wuhan43oo48，China）

Abstract：Whentheexistingdetectionmethodsidentifysmalldefectsofinsulatorsintransmisionlineimageswithcomplex backgrounds，theobtainedimages have problemssuchascomplexbackgroundenvironmentandsmalldefect size.Inorderto ensurethesafeoperationof transmision lines，aninsulatordefectdetectionnetwork（IDD-Net）basedonYOLOv7-tinyis proposed.Theatention-basedintra-scalefeatureinteraction（AIFI）isintroducedtoandleig-dimensionalfeaturesandreduce computationalcomplexity.Thebidirectional weightedpath featurepyramidnetwork（BPFPN）isusedforthefeaturefusion.The improvementsaremadetothedown-samplingmodule，soastoenhancethenetwork'sperceptualcapabilies.TheFocal-DIoU loss function isused toimproveanchorboxquality.Theresultsshowthat，incomparisonwiththebaselinemodel，theaverage accuracyofIDD-Netisimprovedby4.1%，theaccuracyandrecalareimprovedby2.4%and6.5%，thenumberofparameters and floating-point operations are reduced by 5.8% and 2.3% ，respectively，and the average accuracy for flashover defects is improvedby11.2%.Itdemonstratesthattheproposedmethodhassmallerparameters，beterperformanceand stronger robustness.

Keywords：YOLOv7-tiny;insulatordefectdetection;atention-based intra-scalefeatureinteraction;bidiretionalweighted path featurepyramid network;MCdown-samplingmodule;lightweight network

0 引言

随着电网规模的不断扩大，输电线路架设所需的面积也在逐渐增长。（剩余11723字）