基于Y0L0v8智慧城市的小目标检测

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）23-0053-07

Abstract：Toaddressthe problemsofinsuffcient detectionaccuracy，weak featurerepresentation，and highcomputational complexityinsmallObject Detection forsmartcities，this paper proposes alghtweight Object Detection model YOLOv8-CFM improved basedonYOLOv8.The VisDrone2O19 dataset is constructed and cleaned to generate the underlying dataset DataS，so asto enhance thequalityofsmallobjectsamplesandthegeneralizationabilityofthe model.Inthemodeldesign，alightweight backbone network ConvNeXtisintroduced toeduce thenumberof parametersandcomputationalload，improvingadaptability toedgedevices.AFeatureFusionModule thatitegrateslow-leveldetailsandhigh-levelsemantic informationisdesigned，which is combinedwith theMulti-Scale SpatialPyramidAtentionmechanism toenhance theperceptionandlocalizationcapabilitiesof small objects.Experimental results showthat YOLOv8-CFMachieves a mAP @0.5 of 0.640 and mAP@0.5：0.95 of 0.223 on the DataS dataset with only 2.27M parameters and 6.6 GFLOPs. The number of parameters is reduced by 24.6% ，thecomputational load is decreased by 19.5% ，the detection precision is improved by 59.6% ，and the mean Average Precision is increased by 5.4% The results demonstrate thatthe proposed method balances detectionacuracyandeficiencywhilemaintaining lightweight properties， making it suitable for edge computing scenarios in smart cities.

Keywords：YOLOv8;smallobject; smartcity;ObjectDetection

0 引言

随着城市化进程的不断加快，智慧城市建设已成为推动社会治理现代化和城市高效运行的重要方向。（剩余9809字）