航空兵器2025，Vol.32Issue(5)：92-103,12.DOI:10.12132/ISSN.1673-5048.2025.0085
基于多域特征去噪融合网络的无人机目标跟踪算法
UAV Object Tracking Algorithm Based on Multi-Domain Feature Denoising and Fusion Network
摘要
Abstract
To address the issue of performance degradation caused by noise contamination in target features during UAV tracking in complex scenarios,this paper proposes a UAV tracking algorithm based on a multi-domain feature de-noising and fusion network.Firstly,downsampling operation is performed in both the spatial domain and wavelet do-main to reduce noise-contaminated target information.Then,high-resolution features are restored through an upsamp-ling network module in the spatial domain,while the inverse wavelet transform is applied in the wavelet domain for fea-ture reconstruction.Furthermore,a denoising feature fusion module is designed to further purify interference noise,and a downsampling distance loss function is introduced to enhance the denoising effect.Experimental evaluations are con-ducted on four popular UAV benchmark datasets:UAVTrack112,UAVTrack112-L,UAV123,and UAV123@10fps.The results demonstrate that the proposed algorithm has outstanding robustness against interference attributes such as occlusion,scale variation,and appearance change,with an average tracking speed of 127.0 frames per second on an 3090 GPU and 31.1 frames per second on the embedded Jetson AGX platform,enabling effective UAV tracking.关键词
无人机/目标跟踪/多域特征/特征去噪/融合网络/离散小波变换Key words
unmanned aerial vehicle/target tracking/multi-domain feature/feature denoising/fusion network/discrete wavelet transform分类
武器工业引用本文复制引用
王海军,綦丽华,袁伟,郝伟..基于多域特征去噪融合网络的无人机目标跟踪算法[J].航空兵器,2025,32(5):92-103,12.基金项目
国家自然科学基金项目(62103060) （62103060）
山东航空学院研究生创新基金项目(SHSYCX13) （SHSYCX13）
山东航空学院博士启动基金资助项目(2021Y04) （2021Y04）
