文档名：基于多尺度卷积注意力机制的输电线路防振锤缺陷检测
摘要：作为输电线路中的重要金具部件,防振锤的缺陷将对输电线路构成严重威胁.针对由于防振锤缺陷样本数量稀少、背景复杂、区域形状尺寸不一造成的防振锤缺陷识别能力不足的问题,提出一种基于多尺度卷积注意力机制的防振锤缺陷检测方法.首先,通过统计不同缺陷的防振锤尺寸,设计适应不同类别的多尺度卷积注意力机制,使网络重点关注图像中的防振锤区域;其次,引入结构重参数化方法,以将网络中的多分支结构无损失地转换为单分支结构,在提高网络检测性能的同时维持检测速度在较高水平;最后,以渐进式特征金字塔网络结构(AFPN)为基础,融合更多的浅层网络,提高了网络检测防振锤小目标的能力.实际收集的防振锤缺陷数据集实验结果表明,设计的检测方法可显著提升防振锤缺陷检测的性能,检测精度mAP0.5达到了91.9%,在TITANXP平台下检测速度达60.88帧/s,可为输电线路防振锤智能化巡检提供参考.
Abstract：Thepresenceofdefectivedampersinpowertransmissionlinesposesasignificantrisktothesecureandstableoperationoftheelectricalgrid.Advancingtheintelligentdevelopmentofdamperinspectionsintransmissionlines,afastandaccuratedefectdetectionmethodholdsparamountimportance.Addressingtheissueofinsufficientdamperdefectrecognitionduetoscarcedefectsamples,complexbackgrounds,andvaryingregionaldimensions,anoveldamperdefectdetectionnetwork(RCA-YOLOv8)basedonamulti-scaleconvolutionattentionmechanismwasproposed.Firstly,thediversesizesofdampersinimagesareanalyzedandamulti-scaleconvolutionattentionmechanismcomposedofthreesetsofbar-shapedconvolutionsisconstructedtopreciselycapturefeaturesofdifferent-sizeddampers.Subsequently,astructuralreparameterizationmethodisutilizedtoconvertthemulti-branchstructureinthenetworkintoasingle-branchstructure,enablingtomaintainconsistentinferencespeedwiththesingle-branchstructurewhilebenefitingfromthedetectionperformanceimprovementbroughtbythemulti-branchstructure.Inaddition,basedontheYOLOv8featureextractionstructure,aConvBlockstructurecontainingConv-Astructureandstructuralreparameterizationmethodwasconstructedtoproposemulti-scalefeaturesofdampers.Moreover,moreshallownetworkfeaturesareintegratedbyusingtheAFPNsstructure,resolvingfeatureconflictsbetweenlarge,medium,andsmalltargetsintheimages,therebyenablingaccuratedetectionofsmalldampertargetsandfurtherenhancingdetectionperformance.Inthismodel,Conv-Aismoreabletofocusonthedampersareaintheimage,reducingbackgroundinterference,andstructuralreparameterizationgreatlyreducescomputationalcosts.AFPNssolvetheproblemoffeatureconflictsbetweenlargeandsmalldampersintheimage,thusachievingalowcomputationalcostandhighdetectionaccuracymodel.Formodelexperimentation,adatasetofdamperdefectsinpowertransmissionlineswithinsubstationscenesusingimageprocessingtechniquesisgenerated.Toenhancedatasetdiversityandsaveannotationtime,employautomaticdataaugmentationoperations,includingscaletransformations,suchasy-axisflipping,addingGaussiannoise,addingsaltandpeppernoise,adjustingbrightnessandnon-scaletransformations,suchassizeenlargementandreduction,alongwithautomaticallygeneratedannotationfiles.Basedonthedataset,theRCA-YOLOv8networkachievesanaverageprecisionmAP0.5of91.9%andmAP0.5:0.95of77.5%.Comparedwithotheradvancedone-stageandtwo-stageobjectdetectionnetworks,RCA-YOLOv8hasbetterdamperdefectdetectionperformance.ThemAPvaluesofRCA-YOLOv8networkincreasedby5.4%and4.2%respectivelycomparedtothebasicnetworkYOLOv8,withaninferencespeedof60.88framespersecondunderTITANXPplatform.ItcanbeconcludedthattheproposedRCA-YOLOv8networkcanrapidlyandeffectivelydetectdampersandtheirdefectsinpowertransmissionlines.Thefollowingconclusionscanbedrawnfromthesimulationanalysis:(1)Thenetworkbasedonthemulti-scaleconvolutionattentionmechanismcanfocusoncrucialregionsintheimages,suppressingbackgroundregions'featurerepresentationstoobtainmorerelevantinformation.(2)Structuralreparameterizationsuccessfullyconvertsthemulti-branchstructureintoasingle-branchstructurewithoutanyloss,strikingabalancebetweendetectionaccuracyandspeed.(3)AFPNswithprogressivefeaturefusionfromdifferentlevelsenablethenetworktoachievemoreprecisedetectionofsmalldampertargets.
作者：张烨 李博涛 尚景浩 黄新波 翟鹏超 Author：ZhangYe LiBotao ShangJinghao HuangXinbo ZhaiPengchao
作者单位：西安工程大学电子信息学院西安710048西安微电子技术研究所西安710000
刊名：电工技术学报 ISTICEIPKU
Journal：TransactionsofChinaElectrotechnicalSociety
年，卷(期)：2024, 39(11)
分类号：TM755
关键词：防振锤 深度学习 注意力机制 实时缺陷检测
Keywords：Damper deeplearning attentionmechanism real-timedefectdetection
机标分类号：TP391P631.5TP277
在线出版日期：2024年6月18日
基金项目：国家自然科学基金，西安市科技计划，陕西省科学技术协会青年人才托举计划项目，金属成形技术与重型装备全国重点实验室开放课题，西安工程大学博士科研启动基金资助项目基于多尺度卷积注意力机制的输电线路防振锤缺陷检测[
期刊论文] 电工技术学报--2024, 39(11)张烨 李博涛 尚景浩 黄新波 翟鹏超作为输电线路中的重要金具部件,防振锤的缺陷将对输电线路构成严重威胁.针对由于防振锤缺陷样本数量稀少、背景复杂、区域形状尺寸不一造成的防振锤缺陷识别能力不足的问题,提出一种基于多尺度卷积注意力机制的防振锤缺陷...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
基于多尺度卷积注意力机制的输电线路防振锤缺陷检测 Defect Detection of Transmission Line Damper Based on Multi-Scale Convolutional Attention Mechanism
基于多尺度卷积注意力机制的输电线路防振锤缺陷检测.pdf
- 文件大小:
- 3.26 MB
- 下载次数:
- 60
-
高速下载
|
|
