基于改进YOLOv5s的SAR影像铁道检测技术研究

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：2095-2945（2025）13-0050-05

Abstract：Gaofen-SARsatelitesforrailroadinspectionhavetheadvantagesofwidecoverage，all-weatherandmetal sensitivity，butneedtosolvetheproblemofinstantdetectionofrlwaytargets.Forthisreasonanimprovedmodelcalled Lightweight-YOLOv5sisproposedbasedonYOLOv5s，whichisespeciallysuitablefordetectionofrailroadtargetsinSARimage. ByreducingtheMobile-Darknetbackbonefeatureextractionnetwork layers，weoptimizedthenetworkstructure；byaddingHDC andCBMmechanisms，weadjustedthesmalltargetsensoryfieldweightsandstrengthenthesmalltargetlinefacilityfeature extraction;byusing FPGMpruning，weeliminatedtheredundantfeaturemodulesandachievealightweightmodel;byusing VarifocalLossasthelossfunction，weequalizedthepositiveandnegativecategoriesandhighlightthecontributionofpositive examples.The results show that，the accuracy of Lightweight-YOLOv5s model achieves 9 7 . 6 % ，and the inference time reduces to 6.87ms.Comparedwiththeclasicalalgorithmsfordetectinglineartargetsinremotesensingimages，theperformanceisgreatly improved for instant detection of railroad targets.

Keywords： SAR; railroad inspection; rail target detection; YOLOv5s; lightweight network

铁道线路覆盖范围广、距离长，所处地区地形、地貌条件复杂多样，尤其是气候及地质条件复杂的山区，地震、洪水、泥石流等自然灾害易发，会严重威胁铁路系统安全稳定运行，需要即时准确地监测铁路所处环境情况，才能有效保障铁路安全。（剩余5705字）