基于改进RT-DETR模型的油菜田间杂草识别研究

打开文本图片集

关键词：油菜；改进RT-DETR模型；RT-DETR模型；杂草识别

中图分类号：TP391.4；TP183；S451.1 文献标识码：A

文章编号：0439-8114（2025）08-0001-09

DOI：10.14088/j.cnki.issn0439-8114.2025.08.001 开放科学（资源服务）标识码（OSID）：口

Research on rapeseed field weed recognition based on improved RT-DETR model

ZHANGLei，LENGXin，CHENJia-kai，LI Zong-xuan （College of Computerand Control Engineering，Northeast Forestry University，Harbin15OO4O，China）

Abstract：FourtialdXantmtrumaieriaidis，henopoumlbumbosiatemisifoiiniapsdeld weretakenastheresearchobjects.Keychalenges inweeddetection，includingsmallseedlingtargets，weakfeaturesof withered weds，anddificultyinidentifinghighlyoverlappingareas，ereaddressdbyproosinganimproveddetectionmethodbasednthe RT-DETR（Region transformerDETR）model.Theasymptotic eature pyramid network（AFPN）replacedtheoriginalcross-scale contextfusionmodule（CCFM）intheRT-DETRmodel，effectivelyresolvingtheimbalancedfeaturedistributionissueinwitheredwedscausedbyblurredtextureandfeature sparsity.TheSPD-Convmodulewas introduced intothebackbone network toenhance thefeaturerepresentationcapabilityforsmalltargetweds.Theconvolutionalblock atentionmodule（CBAM）wasintegrated attheendofthebackbonenetwork，efectivelymitigatingfeatureinformationlossunderlow-resolutiontargetsandolusionconditions.Systematic ablation experiments and comparative experiments verified that the improved RT-DETR + AFPN + SPD-Conv+CBAM （RW-DETR）modeldemonstratedsignificantadvantagesinbothdetectionaccuracyandrobustnessTheRW-DETRmodelachieved recognition precision and mean average precision of 85.2% and 82.5% ，respectively，for weeds in rapeseed fields，significantly outperformingtheRT-DETRmodel，FasterR-CNNmodel，SSDmodel，YOLOv5m model，andYOLOv8m model.Whilemaintainingrealtimedetectionperformance，theRW-DETRmodelsignificantlyimprovedweedrecogitionefectivenessincomplexsenes，mting theaccuracy and efficiency requirements of modern agriculture for weed detection systems.

Key Words：rapeseed;improved RT-DETR model; RT-DETR model;weed recognition

油菜是全球种植广泛的油料作物，油菜子主要用于榨取菜子油。（剩余12134字）