邊緣計(jì)算下的絕緣子缺陷小樣本檢測研究

打開文本圖片集

引用格式：，等.邊緣計(jì)算下的絕緣子缺陷小樣本檢測研究[J].現(xiàn)代電子技術(shù)，2025，48（10）：76-84.

關(guān)鍵詞：絕緣子缺陷；小樣本檢測；RT-DETR編碼器；邊緣計(jì)算；距離嵌入模塊；路徑聚合網(wǎng)絡(luò)中圖分類號：TN919.8-34；TP391.41 文獻(xiàn)標(biāo)識(shí)碼：A 文章編號：1004-373X（2025）10-0076-09

Abstract：Inorder tosolvetheproblemsoflowprecisionand poorrobustnessof traditionaltargetdetectionalgorithms when detectingsmall-sampleinsulatordefetsontransmissionlines，andtorealizetheeficiencyofUAVpatrolinspection，anedge computing based featuredistacediferencesmal-sampleinsulatorself-explosion detectionalgorithm is proposed.TheRT-DETR encoderisimprovedbymeansofhighandlowfrequencyinformationfusion（AHiLo）and（hierarchicalscale-basedpath aggregationnetwork，HS-PAN）toextractlocalhighandlowfrequencyinformationof insulatorstrings.AdistancedierenceXgoal （DX）isintroduced toobtain theoptimal metricdistancebetweentheprototypeagentandthequeryfeatureinthemapping featurespace，soastorealizetheaccuratedetectionofsmall-sampleinsulatorself-destruction.Theexperimentalresultsshow thattheimproved modelcanrealizeadetectionacuracyof 86.4%usingonly15Osampleson thePCside，with a parameter count of 2.06×107 ，and adetection speed of 66.1f/s，which meets the requirements of smal-sample detection.In comparison with othermainstreamalgorithms，theimprovedalgorithmcanshowahighlevelof detectionaccuracyandreal-timeperformance.

Keywords：insulator defect;small-sampledetection;RT-DETRcoder;edgecomputing;DX module;pathaggregation network

0 引言

《2022年全國電力可靠性年度報(bào)告》中指出，在 220kV 及以上架空線路的非計(jì)劃停運(yùn)原因中，導(dǎo)線、絕緣子和金具位列前三，其中絕緣子引起非計(jì)劃停運(yùn)49次，引起非計(jì)劃停運(yùn)時(shí)間為 1 287.05h 。（剩余10005字）