基于低轨卫星联合监视的机场飞机目标跟踪方法
Airport Aircraft Target Tracking Method Based on Joint Surveillance of Low Earth Orbit Satellites
陈家建 1刘勇 1郭鹏宇 1曹璐 1王鑫慧 1孟玲1
作者信息
- 1. 军事科学院国防科技创新研究院,北京 100071
- 折叠
摘要
随着大规模低轨商业遥感卫星星座的发展,卫星能够在较短时间内对地面同一区域进行接力观测,使得对机场飞机目标的持续监视成为可能.针对低轨卫星联合监视下的机场飞机目标跟踪问题,文章提出了一个基于YOLOv7 与DeepSORT的两阶段飞机目标跟踪方法,通过引入注意力机制改进YOLOv7 网络,实现对卫星遥感图像中小型飞机目标的准确检测;通过优化匹配机制改进DeepSORT算法,实现对存在长时间差的图像序列中运动飞机目标的持续跟踪.文章利用多颗"吉林一号"卫星接力拍摄的卫星遥感图像序列进行算法验证,实验结果表明,文章方法能够融合多颗卫星不同时刻的成像信息,提高对机场运行状态的连续感知能力,多目标跟踪精度为 79.15%,对比其他常用的跟踪算法提高了3.2%以上,同时显示出大规模低轨商业遥感卫星星座在目标跟踪领域巨大的应用潜力.
Abstract
With the development of large-scale commercial remote sensing Low Earth Orbit Satellite Constellations,satellites are capable to work in relay observation mode over the same ground area within a shorter period,which enables continuous monitoring of aircraft targets at airports.This paper proposes a two-stage aircraft target tracking method based on YOLOv7 and DeepSORT for airport aircraft target tracking under joint monitoring of low orbit satellites.By introducing an attention mechanism to improve the YOLOv7 network,accurate detection of small aircraft targets in satellite remote sensing images is achieved;By optimizing the matching mechanism and improving the DeepSORT algorithm,continuous tracking of moving aircraft targets in image sequences with long time differences can be achieved.This article uses satellite images sequences captured by multiple Jilin-1 satellites for algorithm validation.The experimental results show that the method proposed in this paper can integrate imaging information from multiple satellites at different times,improve the continuous perception ability of airport operation status,and achieve a multi-target tracking accuracy of 79.15%,which is more than 3.2%higher than other commonly used tracking algorithms.At the same time,it shows the significant application potential of large-scale low-Earth orbit commercial remote sensing satellite constellations in the field of objects tracking.
关键词
低轨卫星星座/多目标跟踪/改进DeepSORT算法/注意力机制Key words
low earth orbit satellite constellation/multi-object tracking/improved DeepSORT algorithm/attention mechanism引用本文复制引用
出版年
2024
NETL
NSTL
万方数据