基于跨分量协同融合与多阶非局部通道注意力的绝缘子缺陷检测方法

打开文本图片集

关键词：绝缘子；缺陷检测；跨分量协同融合模块；多阶非局部通道注意力；特征判别；频域信息中图分类号：TN911.73-34；TM216；TP391.41 文献标识码：A 文章编号：1004-373X（2026）06-0160-08

Method of insulator defect detection based on cross-component collaborative fusion and multi-order non-local channel attention

TANGYifan²，YUMei²，LULin，²

HubeiKeybeldoei 2.CollegeofComputerandInformationTechnology，ChinaThreeGorgesUnversity，Yichang443oO2，China）

Abstract：Inalusion tothehigh similaritybetwee targetsandbackgroundininsulatorimages，aswellasthedefect featuresofsmalltargetsarepronetoeingdilutedduetodownsamplingandlimitedreceptivefields，aninsulatordefectdetection algorithmbasedonc-ompontcollaborativefusiadultidero-localcaelaetiisproosd.Ao-oot colaborativefusionmoduleisintegatedintothebackbonenetwork，andthecrossdomainfusionisconductedbymeansoffrequencydomainandspatialdomainfeature，torealizethemulti-scalecandenhancethefeaturediscriminationability，therebyimprovingtherecognitionefectofsubtledefectdiferences.Inthenecknetwork，amulti-ordernon-localchanelatentionmechanismisintroducedtocaptureinter-chanelcorelationsatmultiplesales.Inombinationwithon-localperception，ithnces therepresentationofsmalldefectregions，suppreses feature dilutioncausedbydownsampling，and thenimprovesdetection accuracy for small-scale defects.The experimental results show that the improved model can realize mAP @0.5 and mAP @ 0.5：0.95 of 79.7% and 38.6%，respectively，which are 3.6%and 3.8% higher than thoseof YOLOv8 benchmark model.The AP in the insulator defect category can reach 79.3% ，and the frame rate can reach 6O.2 f/s，which can meet the real-time detection requirements of the power system and significantly improve the detection accuracy of insulator defects.

Keywors：insulator;defectdetection;cross-componentcollaborativeinteractionmodel；multi-ordernon-localchanne attentionmechanism;featurediscrimination;frequencydomaininformation

随着全球电力需求的不断增长和电力传输网络的扩展，电网系统的稳定性和可靠性面临严峻挑战。（剩余12580字）