基于改进YOL011的遥感图像军事伪装目标识别算法研究
打开文本图片集
关键词：遥感图像；伪装目标；双向尺度融合；空洞卷积；感受野；通道位置注意力中图分类号：TN911.73-34；TP391.4 文献标识码：A 文章编号：1004-373X（2026）09-0042-09
Improved YOLO11 based military camouflage target recognition algorithm for remote sensing images
Li Xiaoqing，Yang Nan （RocketForceUniversityofEngineering，Xi'an71oo25，China）
Abstract：Inviewof thecomplexbackgrounds，extremelysmalltarget scales，lowresolution，and theconfusionbetweenthe targetforegroundandbackgroundoftheremotesensing images，theauthorproposestheRSC-YOLOalgorithm.Onthebasisof YOLO11algorithm，thebackbonenetwork isfirstimproved.Thespatial-to-depthconvolution （SPDConv）isused toreplacethe downsampling，whichavoidsthefine-grainedinformationlosscausedbythetraditionalstrideconvolutionandpoolingoerations， soastolearnmoreeffectivefeaturerepresentationsoflow-resolutionimages.Secondly，abi-drectionalsalesequencefeature fusion（BSSFF）moduleisdesignedintheNecklayerandabi-directionalinformationinteractionmechanismisintroduced， enablingmoresuffcient information exchangeamong feature mapsof different scalesandbeteradaption tothediversityand complexityofcamouflagedtargetsinremotesensingimages.Furthermore，adilatedconvolutionbasedchannelandposition atentionmechanism（DCPAM）moduleisdesigned，whichcanexpandthereeptivefieldwithout increasingparametersand computationalload.Whencalculatingthepositionatentionweights，themodelcanhighlightthekeypositionsofcamouflaged targets，suppessbackgroundnoiseinterference，andeffectivelyenhanceitsabilitytocapturecontextualinformation.all，a datasetofcamouflagedtargetsinremote-sensingimages isconstructedfortrainingandtesting.Theexperimentsindicatethat，in comparison withthebaselineYOLO1lmethod，theproposedRSC-YOLOalgorithm，withonlyamarginal increaseoftheparameters， achieved an 11.2% improvement in Fβω and a 31.1% decrease in MAE （mean absolute error）， respectively，and could reach a processingspeedof35f/s（framespersecond）.Ablationexperimentsdemonstratedthatthetheedesigedmodulesouldffectively enhance theperformanceofthealgorithmandfulfilltherequirementsforeal-timedeploymentonsmall-scaleairbornedevices.
Keywords：remotesensingimage;camouflaged target;bi-directional scale fusion；dilatedconvolution；receptive field; channel and position attention
0 引言
随着现代军事技术与侦察手段的飞速发展，遥感图像在众多领域尤其是军事国防和国土安全方面发挥着极为关键的作用。（剩余12234字）