基于UAV-YOLO的无人机航拍图像轻量化目标检测算法

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（2025）15-0051-06

UAV-YOLO-based lightweight object detection algorithm for UAV aerial images

1，2，3， 1，2，3， 1，2，3 （1.Collegeof Computer and Cyber Security，Hebei Normal University，Shijiazhuang O5oO24，China; 2.Hebei KeyLaboratory of Network and Information Security，Shijiazhuang O5oo24，China; 3.HebeiProvincialEngineeringResearchCenterforSupplyhainBigDataAnalyticsandDataSecurityShijiazuangO524ina）

Abstract：Inresponse to thechallenges inunmannedaerial vehicle（UAV）imageanalysissuch ascomplex backgrounds， lowaccuracyindetectingsallobjectsandighmissinginspectionrates，a UAV-YOLO，lightweightobjectdetectionalgorithm basedoYOLOv7specificallydesignedforUAVaerialimages，isproposed.Firstly，alightweightconvolutionmodule，PSConv （partialconvolution），isconstructedbasedonpartialconvolutionstoreducenetworkcomplexitywhilemaintainingdetection performance.Secondly，a downsampling module is integratedwithanatention mechanism to buildtheMA-ECAmodule.Then， thedetectionheadsofthenetworkisoptimizedbyaddingsmallobjectdetectionheadsandremovinglargeobjectdetection heads.Finallytheocal-SIoUlossfunctionisproposedtofurtherimprovethedetectionaccracyofthemodel.Thealgorithm wasvalidatedonthepubliclyavailableVisDrone2O19andUAVDTdatasets.Incomparisonwiththeother models，theproposed algorithm improves the detection accuracy efectively while reducing network parameters and computational complexity.

Keywords：YOLOv7; UAV;smallobject detection;lightweight network;partial convolution;attntion mechanism

0 引言

近年来，随着无人机技术越来越成熟，无人机被广泛应用于各个领域，如智慧城市、灾后救援和农业检测[2。（剩余8284字）