基于改进YOLOv10的全天候无人机航拍图像检测算法

打开文本图片集

中图分类号：TN911.73-34 文献标识码：A 文章编号：1004-373X（2026）07-0048-07

Abstract：Inviewofthesinglesceneandlowdetectionaccuracyof theexistingUAVaerialimagedetectionalgorithms， thispaper proposesanall-weather UAVaerial image detection algorithm FEMFF-YOLO basedonthe improved YOLOv10. Firstlyanal-weatherUAVaerialimagedatasetDNV（day-night-visible）isconstructedtoevaluatetherobustnessand generalizationabilityoftheobjectdetectionmodelduring multipletimeperiodsandmultipleweatherconditions.Secondly，the originalmodelbackboneiseplacedwiththeFasterNetstructurethatcanextractfeaturesmoreeficiently.Finally，RepHMS，a re-parameterized heterogeneousmulti-scalemodule，isusedtoenhancethemulti-scalefeaturefusionabilityofthemodel. Experimentsshow that the accuracy of FEMFF-YOLO algorithm is improved by 72.8% ， its mean average precision mAP@0.5 reaches 63.8% ，andits recallratereaches 57.9%on the dataset DNV.In comparisonwith those of the basic YOLOv10 algorithm，itsaccuracyisincreasedby3.9%，itsmAP by 3.3% ，anditsrecallrateby2.3%.TheeffectivenessofUAVaerial target detection is verified.

Keywords：deep learning;UAV;aerial image;YOLOv1O;all-weather; feature extraction

0 引言

如今，无人机航拍技术已广泛应用于军事侦察、灾害救援、农业监测、城市规划、交通管理等领域，产生了海量的低空遥感图像数据。（剩余11249字）