基于图注意力机制的无人机目标红外跟踪技术  

Infrared UAV target tracking technology based on graph attention mechanism

作　　者：师晓冉 张焱[2] SHI Xiaoran;ZHANG Yan(Unit 93202,Urumqi 841700,China;College of Electronic Science and Technology,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]93202部队,新疆乌鲁木齐841700 [2]国防科技大学电子科学学院,湖南长沙410073

出　　处：《国防科技》2023年第3期26-39,共14页National Defense Technology

基　　金：国家自然科学基金项目(62075239、61302145)。

摘　　要：近些年来,无人飞行器逐步普及,给社会治安、空域管制、要地防护等领域带来新的威胁。因此,对高威胁、难辨识的无人机目标进行高准确、高稳定的搜索跟踪有着迫切需求。立足城市环境下的反无人机作战需求,针对无人机易受复杂背景干扰、地物遮挡影响及其姿态尺度变化和快速机动造成的跟踪丢失问题,结合图神经网络与注意力机制,提出一种基于图注意力机制的孪生网络无人机目标红外跟踪技术。基于图注意力机制的局部特征匹配模块,通过节点间的信息传递实现全局信息匹配,通过基于图的特征表示和特征交互,增强从目标模板到搜索区域的信息嵌入能力;目标感知区域选择机制可以适应无人机目标尺寸和姿态的变化;基于运动特征设计的切换策略,实现局部跟踪和全局重检测间的动态选择。实验结果表明,在红外无人机目标跟踪时,运用此方法可以取得稳定的跟踪结果,可以有效应对形变、遮挡、复杂背景干扰等挑战,为实现在城市环境中持续稳定跟踪小型无人机目标提供技术参考。In recent years,the increasing popularity of Unmanned Aerial Vehicles(UAVs)has brought new threats to fields such as public security,airspace control,and the protection of prominent locations.Therefore,there is an urgent need for highly accurate and stable tracking of UAVs,which are exceedingly threatening and difficult to identify.This study is based on the requirements of anti-UAV combat in urban environments.As UAVs are susceptible to complex background interference,object occlusion on the ground,attitude scale changes,and rapid maneuvering,they cause problems in tracking.To address this issue,a Siamese network with infrared technology is proposed,whereby a graph neural network is combined with an attention mechanism to track UAVs.The local feature matching module that is based on the graph attention mechanism achieves global information matching through information transmission between nodes and enhances information embedding from target template to search area through graph-based feature representation and feature interaction.The selection mechanism for intercepting the target region can adapt to the changes in target size and the attitude of UAV.A switching strategy based on motion features is designed to achieve dynamic selection between local tracking and global re-detection.The experimental results show that the proposed method achieves stable infrared UAV target tracking and can effectively deal with challenges such as deformation,occlusion,and complex background interference,thereby providing a technical reference for achieving continuous and stable tracking of small UAV targets in urban environments.

关 键 词：反无人机 红外目标跟踪 孪生网络 图注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]