一种基于地基望远镜数据的目标检测方法  

A target detection method based on ground-based telescope data

作　　者：杨笑天 李伟[1,2] 阴晓刚 王梁 谭金林 吴意 曹小敏 YANG Xiaotian;LI Wei;YIN Xiaogang;WANG Liang;TAN Jinlin;WU Yi;CAO Xiaomin(China Academy of Space Technology(Xi′an),Xi′an,Shaanxi 710100,China;Shaanxi Aerospace Technology Application Research Institute Company Limited,Xi′an,Shaanxi 710100,China)

机构地区：[1]西安空间无线电技术研究所,陕西西安710100 [2]陕西航天技术应用研究院有限公司,陕西西安710100

出　　处：《北京测绘》2024年第12期1754-1759,共6页Beijing Surveying and Mapping

摘　　要：本文对地基望远镜序列图像进行不均匀背景校正、快速配准、空间卫星目标检测,在此基础上按时间序列对空间卫星目标进行正向目标关联识别,提取移动目标的运动轨迹,剔除伪目标点。在序列图像帧间关联过程中提出了剪枝的策略,对错误轨迹进行删除,避免了在搜索过程中出现的维数爆炸式增长问题,并对缺帧情况下的轨迹中断问题采用了反向关联的思想,在正向关联完成后,再按反向时间序列进行帧间目标关联识别。最后再把每一帧的正向关联识别结果和反向关联识别结果合并,实现对地基望远镜序列图像中目标的关联识别,降低了虚警率、提升了检测速率。结果表明:在虚警点数较多时,帧数越多检测越准确,目标数量越多速度优化越明显。通过试验证明,该方法检测结果与已知的标签结果一致。For ground-based telescope sequence images,this paper performed uneven background correction,fast image registration,and spatial satellite target detection.On this basis,the paper performed forward target association recognition on spatial satellite targets according to time series,extracted the motion trajectory of moving targets,and removed pseudo target points.A pruning strategy was proposed in the inter-frame association process of sequence images to delete erroneous trajectories,avoiding the problem of explosive dimensionality growth during the search process.The idea of reverse association was adopted for trajectory interruption in the case of missing frames.After the forward association was completed,the inter-frame target association recognition was carried out according to the reverse time series.Finally,the forward association recognition results and reverse association recognition results of each frame were merged to achieve association recognition of targets in ground-based telescope sequence images,reducing false alarm rates and improving detection rates.The results show that when there are many false alarm points,more frames indicate more accurate detection,and more targets contribute to obvious speed optimization.Through experimental verification,the detection results of this method are consistent with the known label results.

关 键 词：背景不均匀校正 图像配准 图像去噪 目标检测 帧间关联 

分 类 号：P237.2[天文地球—摄影测量与遥感]