改进YOLOv8n的无人机航拍目标检测算法

打开文本图片集

引用格式：.改进YOLOv8n的无人机航拍目标检测算法[J].现代电子技术，2026，49（1）：77-85.

关键词： ΥOLOv8n ；密集目标检测；C2f-RFCBAMConv；跨尺度连接；特征融合；注意力机制中图分类号：TN911.73-34；TP391 文献标识码：A 文章编号：1004-373X（2026）01-0077-09

UAV aerial photography object detection algorithm based on improved YOLOv8n

QU Wenlong， CHEN Yong （School of Mathematicsand ComputerScience，Shaanxi Universityof Technology，Hanzhong 723ool，China）

Abstract：In viewof the missed detection causedby multi-scaleandoclusion in dense object detection tasks，this paper proposesanalgorithmonthebasisof theimproved YOLOv8n，andthealgorithm isnamedasDERF-YOLO.Inthebackbone network，theC2f-RFCBAMConvmoduleisused toreplaceconvolutionandC2f modules，andthereceptivefieldattention mechanismisutilizedtooptimizespatialfeaturesandtheniscombinedwithchannelatention mechanisms toenhance feature extractioncapabilitiesoftheimprovedalgorithm.TheEMBSFPmoduleisdesignedtooptimizetheneckpart，ahievingrossscaleconnectionandfeature fusionofsmallobjectinformation.Byreplacing theup-sampling module，thealgorithmcan maintain eficiencywhilensuringcertainefects.inally，DyHeadisemployedinsteadofteoriginaldetectionhad，andtetetion mechanismsareintroducedtoimprovetheacuracyofsmallobjectdetection.ExperimentalresultsontheVisDrone2Ol9dataset show thatthe mAP@0.5andmAP@0.5：0.95of the DERF-YOLO reach 30.9%and17.7%，respectively，representing improvements of 4.0% and 2.7% ，respectively，in comparison with the YOLOv8n algorithm.The parameter count and floatingpoint operations of the DERF-YOLO are 2.94×106 and 9.6 GFLOPs，respectively， with a 2% reduction in parameters and only an 18% increaseincomputationburdenincomparisonwith thoseoftheoriginalalgorithm.Itsaccuracyishigherthanthatof the othersimilaralgorithmsanditmetsthemonitoringrequirementsandcanbeefectivelyappliedtotheobjectdetectiontaskson UAVaerial photography platform.

Keywords：YOLOv8n；denseobject detection;C2f-RFCBAMConv;cross-scale connection;featurefusion；attention mechanism

0 引言

在航拍场景中，如何准确、快速地检测到小目标是一个重要的问题，但是由于航拍高度较高，导致地面目标的尺寸相对较小，目标在图像中的像素占比低，传统检测方法难以从背景中分离出这些目标。（剩余9251字）