改进YOLOv8的轻量级无人机跟踪方法

打开文本图片集

中图分类号：TP391.41 文献标志码：A DOI： 10.12305/j.issn.1001-506X.2026.03.01

Abstract：At present，the existing unmanned aerial vehicle（UAV） tracking methods have problems such as low detectionaccuracyforlong-range UAVs，large parameter quantities that are dificult to track inreal time，and easytarget loss.Therefore， a lightweight UAV tracking method based on improved you only look once version 8（YOLOv8）is proposed. In response to the problem low detection acuracy long-range UAVs using existing methods，YOLOv8 is used as the baseline model to replace the originalconvolution module in the network structure with spatial to deep grouped convolution，which improves the model's feature extractionabilityforsmal targetswhilereducing network parameters.To addressthe problem dificulty inreal-time tracking due to the large number model parameters，a dep separable shuffle network structure is designedas thebackbone network the model，which reduces thenumber model parameters while ensuring detection accuracy.To addressthe issue tracking loss inordinarytracking models，an improved detection model combined with ByteTrack algorithmisused toenhance the tracking performanceUAVs in complex environments.The tracking methodis validated on theReal World dataset，and compared to the baseline model，the improved UAVdetection model shows a 1.6% increase in detection accuracy，a 0.8% increase in recall，a O.2 increase in Fl metric value，a 0.5% （20 increase in average detection accuracy，and a 0.2×106 reduction in parameter count， demonstrating that the model has good detectionacuracyandreal-timeperformance.Tracking testisconductedon UAV flightvideos，and theresultsshowthatthe

proposed method has good performance in UAV tracking.

Keywords ： deep learning；target detection；target tracking；deep separable network structure；you only look once （YOLO）

0引言

近年来，随着无人机飞行技术自主化和智能化的发展，无人机被广泛应用于各个领域。（剩余18088字）