文档摘要：输电线路在跨越高速铁路、高速公路和重要输电通道场景下易受到外力破坏,可能严重影响输电线路安全可靠运行.针对此问题,该文通过构建输电线路走廊隐患目标数据集,提出新模型YOLO-2MCS用于输电线路走廊隐患目标检测.使用混合数据增强策略对数据集进行有效扩充,以提高模型在复杂场景下的泛化性和鲁棒性;在EfficientRep骨干网络引入卷积注意力机制模块,有效提升模型对多尺度目标的检测能力;构建使用softplus激活函数的双向特征金字塔结构加强模型特征学习能力;在检测头使用SIoU损失函数进一步提升模型检测精度.实验结果表明,相较于原YOLOv6网络,该模型在0.5∶0.95的严苛阈值下平均精度均值提升4.4％;将该模型与主流的检测模型FasterR-CNN、YOLOX、YOLOv5和YOLOv7分别进行对比评估,该模型的检测精度、检测速度、模型复杂度均获得最优性能,其平均检测速度高达约300帧/s,且内存仅为40.7MB,同时满足在边缘计算设备上部署的要求.
Abstract：Transmissionlinesarevulnerabletoexternalbreakagewhencrossinghigh-speedrailways,highwaysandimportanttransmissionchannels(referredtoas"Three-Span").Theoccurrenceofsuchaccidentsishighlyrandom,andmaycausesdamagetotransmissionequipment,powergridtripping,poweroutages,andevenaccidentalelectricshockandothersafetyaccidents,whichseriouslyaffectthesafe,reliableandstableoperationoftransmissionline.Theexistingsuper-high/UHVtransmissionlineinspectionmethodsincludemanualinspection,robotinspection,helicopterinspectionandUAVinspection,etc.,andgraduallyformthemaintenanceandoperationmodewithUAVinspectionasthemainandmanualinspectionasthesupplement.However,UAVinspectionalsohaslimitationssuchasdifficultoperation,limitedflightdistance,andmanyuncertaintiesinthefield,whichisnotsuitableforlarge-scalepromotion.Aimingatthelimitationsoftheexistingtransmissionlineinspectionmethodsandthelackoftheexternalforcedamageobjectdataset,thispapercollectedandconstructedahiddendangerobjectdatasetoftransmissionlinecorridor,whichcontainsthreetypicalhiddendangerobjectcategoriesand8654targetsintotal.Thehybriddataaugmentationstrategyisusedtoeffectivelyenrichthedatasettoimprovethegeneralizationandrobustnessofthemodelincomplexscenariosandavoidthemodeloverfittingproblemduringtrainingcausedbyasinglescenario.Theconvolutionalblockattentionmodule(CBAM)isintroducedintheEfficientRepbackbonenetworktoreducethefeaturelossintheextractionprocessoftheoriginalbackbonenetworkandimprovethemodel'sabilitytoidentifyandlocateoccludedobjects.Thebidirectionalfeaturepyramidnetworkusingthesoftplusactivationfunctionenhancethefeaturelearningabilityofthemodelandtheconvergencespeedduringtraining.TheSIoUlossfunctionthatdefinestheanglecostfunctionisintroducedinthedetector,whichcanmakethelossfunctioninthetrainingprocessconvergeassoonaspossibleandcompletetheparameteradjustmentofbackpropagation,sothatthemodelcanobtainbetterandfasterobjectpositioningperformance.Theresultsofablationexperimentsshowedthatthestrategiesproposedinthispapercansignificantlyimprovethedetectionaccuracyanddetectionspeedofthemodelwithoutundulyaffectingthecomplexityofthemodelinthehiddendangerobjectdatasetoftransmissionlinecorridorconstructedinthispaper.ComparedwiththeoriginalYOLOv6model,theaverageaccuracyofthenewmodelisimprovedby4.4％underthestrictthresholdof0.5∶0.95,theaveragedetectionspeedisashighasabout300framespersecond,andthememorysizeisonly40.7MB.TheresultsofcomparativeexperimentsshowedthatthedetectionaccuracyanddetectionspeedoftheproposedmethodaresignificantlybetterthantheFasterR-CNN,YOLOX,YOLOv5andYOLOv7underthesamehyperparametersandthesamedataset.Thevisualizationresultsshowedthatwhenthereisahiddendangerobjectthatareoccluded,theproposedmodelcanalsobetteridentifyandlocateit.Basedontheaboveexperimentalresults,theYOLO-2MCSmodelproposedinthispapercannotonlyaccuratelyidentifythecategoriesofexternalforcedamageinthetransmissionlinecorridormonitoringscene,butalsoquicklyandaccuratelylocatethehiddendangerobject,allwhilemeetingtheneedsofdevicesinstalledonthemobileedge.Thehiddendangerobjectdatasetoftransmissionlinecorridorconstructedinthispaperprovideseffectivesupportforsubsequentmodeltraininganditeration,andeffectivelydevelopstheintelligentdevelopmentofthepreventionofexternalforcedamageintransmissionlinecorridor.
作者：郑含博 胡思佳 梁炎燊 黄俊杰 汪涛 Author：ZhengHanbo HuSijia LiangYanshen HuangJunjie WangTao
作者单位：广西电力系统最优化与节能技术重点实验室(广西大学)南宁530004国网湖北省电力公司电力科学研究院武汉430077
刊名：电工技术学报 ISTICEIPKU
Journal：TransactionsofChinaElectrotechnicalSociety
年，卷(期)：2024, 39(13)
分类号：TM85
关键词：输电线路走廊 防外破 目标检测 注意力机制
Keywords：Transmissionlinecorridor preventionofexternalforcedamage objectdetection attentionmechanism
机标分类号：TP391.41TM732TS207.3
在线出版日期：2024年7月22日
基金项目：国家自然科学基金，国家自然科学基金，广西科技基地和人才专项资助项目基于YOLO-2MCS的输电线路走廊隐患目标检测方法[
期刊论文] 电工技术学报--2024, 39(13)郑含博 胡思佳 梁炎燊 黄俊杰 汪涛输电线路在跨越高速铁路、高速公路和重要输电通道场景下易受到外力破坏,可能严重影响输电线路安全可靠运行.针对此问题,该文通过构建输电线路走廊隐患目标数据集,提出新模型YOLO-2MCS用于输电线路走廊隐患目标检测.使用...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
关键词：输电线路走廊,防外破,目标检测,注意力机制,
- 文件大小:
- 14.98 MB
- 下载次数:
- 60
-
高速下载
|
|
1
