【作者】 龙云利；

【导师】 徐晖；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2012， 博士

【摘要】 天基红外目标监视系统综合了天基平台和红外传感器两者的优势，具有探测距离远、覆盖范围广、测量精度高、隐蔽性强等优点，已成为空间目标跟踪监视的主要途径。本文针对天基红外目标监视系统中的单传感器像平面目标轨迹检测与跟踪问题进行研究，重点探讨了探测图像的非均匀性校正与背景杂波抑制、像平面多目标数据关联与轨迹跟踪、弱小多目标轨迹检测前跟踪等关键技术。论文的主要工作如下：第二章探讨了探测图像的非均匀性校正与背景杂波抑制问题。首先，针对探测图像的非均匀性校正问题，提出了一种改进的单帧场景类神经网络快速校正算法。仿真实验结果表明，所提算法相对传统的神经网络校正等算法具有更好的校正精度和收敛速度。其次，针对高轨静止平台探测中的背景杂波抑制问题进行研究，提出了基于Markov自回归模型的时空域融合抑制算法(Markov-SPAT)和基于约束序贯M估计的时空域融合抑制算法(M-SPAT)。仿真实验结果表明，所提算法相对艾斯卡尔等人的时空分级融合滤波、Tartakovsky等人的时空域融合滤波等算法，在背景杂波抑制、目标信号保持和信噪比改善等方面得到了改善。最后，针对中低轨运动平台探测中的背景杂波抑制问题进行研究，探讨了基于MHD-PSO特征图像配准的时空域融合抑制算法和基于平台运动参数的时空域融合抑制算法。仿真实验结果表明，所提两种算法优于已有的空域滤波等算法。相比而言，基于平台运动参数的时空域融合抑制算法具有更好的实时性。本章为后续章节的研究提供了支撑。第三章研究了高信噪比条件下像平面多目标数据关联与轨迹跟踪问题。首先，针对推扫型和锥扫型传感器两种不同工作模式，建立了像平面目标轨迹的运动模型和观测模型。其次，针对推扫型传感器重点研究了基于重要度采样-马尔可夫链蒙特卡洛(IS-MCMC)算法的多目标数据关联与跟踪技术，实现了目标轨迹的存在性检测和维持跟踪。通过仿真实验对算法的有效性进行了验证，并对算法性能进行了对比分析。仿真实验结果表明，相比于MHT、MCMC、SMC-PHD等算法，所提的IS-MCMC算法具有更好的数据关联和目标轨迹检测跟踪性能。最后，针对锥扫型传感器非线性观测环境，提出了基于扩展无味卡尔曼滤波与改进型积分概率多假设跟踪(AUKF-OIPMHT)的多目标数据关联与跟踪算法。进一步针对像平面非线性运动目标，结合交互多模型(IMM)方法，提出了IMM-OIPMHT算法。仿真实验结果表明，相比IS-MCMC、IPMHT等算法，IMM-OIPMHT具有更稳健的跟踪性能。第四章研究了低信噪比条件下像平面弱小目标轨迹的检测前跟踪(TBD)问题。首先，建立了弱小目标检测前跟踪处理中的像平面目标运动与观测模型。其次，对单目标多模型粒子滤波检测前跟踪(MM-PF TBD)算法进行了优化，并提出了启发式MM-PF TBD算法，将MM-PF扩展应用到未知目标数目、目标出现与消失时刻的天基红外目标监视系统中。仿真实验结果表明，所提启发式MM-PF算法相比于MM-PF、直方图概率多假设跟踪(H-PMHT)、动态规划(DPA)等已有TBD算法具有更优的检测跟踪性能。最后，为解决多目标环境中启发式MM-PF算法由于迭代搜索导致的计算效率下降问题，尝试将近年兴起的概率假设密度检测前跟踪算法(PHD TBD)引入到本文中。针对天基红外监视系统中的像平面非线性运动弱小目标轨迹的检测跟踪问题，在对PHD进行改进的基础上，结合多模型思想，提出了MM-PHD TBD算法。仿真实验结果表明，所提MM-PHD算法与启发式MM-PF算法性能相当，均优于PHD、H-PMHT、DPA等TBD算法。更多还原

【Abstract】 Space-based infrared surveillance system has both the advantages of space-basedobserver and infrared sensor in many aspects, such as far sensing distance, largecoverage area, high measurement precision, concealment ability and so on. Space-basedinfrared surveillance system has been becoming the dominant measurement to keepstrategic targets under surveillance. The dissertation focuses on the problem of detectingand tracking targets in the focal plane array for the sensors of the space-based infraredsurveillance system. The key technologies are discussed and researched especiallyincluding the correction of the images non-uniformity, the suppression for thebackground clutter, the data association and tracking for multiple targets andtrack-before-detect (TBD) for dim targets at low signal to noise ratio(SNR). The maincontributions of this dissertation are demonstrated as follows:In chapter2, the studies are conducted on the correction of the imagesnon-uniformity and the suppression of the background clutter. Firstly, an improvedmethod based on Nerve-Network (NN) using the single scene image is proposed forcorrecting the non-uniformity. As the extensive experiments show that the proposedmethod achieves better performance and converges faster than the traditional NNmethod. Secondly, studies on the suppression of the background clutter for thegeo-stationary satellite is done with the two spatial-temporal fused filtering methodsconsequently proposed respectively based on Markov automatic regression(AR) modeland restricted sequential M-estimation. The experiments validate the superiority of theproposed method in the suppression of the clutter, the maintenance of the target signaland the improvement of the SNR over the being united and non-united spatial-temporalfused methods. Lastly, the studies are implemented on the suppression of thebackground clutter for the moving satellite sensors with low orbit. Two stagedspatial-temporal fused methods are discussed respectively on the images registration byfeatures and on the parameters of the plane with experiments following to compare andanalysis their performance. The results show that the presented algorithms outperformthe spatial filtering ones. The aforementioned researches can serve as the foundation forthe succeeding chapters.In chapter3, the issue of the dada association and multi-target tracking is studied.Firstly, the measurement model and target dynamic model on focal plane are bothestablished respectively for the broom-scanning and cone-scanning sensors. Secondly,for the broom-scanning sensor studies are focused on the multi-target data associationand tracking based on Markov chain Monte Carlo with importance sampling(IS-MCMC) to judge the target existence and then track it. The algorithm is validatedby experiments. Compared to the traditional methods such as MHT, MCMC and SMC-PHD, IS-MCMC has the superiority in data association and target detection. astly,to track multi-target under nonlinear observation for the cone-scanning sensor, a methodfor data association and tracking is proposed on augmented unscented Kalman filter(AUKF) and optimized integrated probability multi-hypothesis tracking (OIPMHT).Further OIPMHT is combined with interactive multiple model (IMM-OIPMHT) totrack the target with nonlinear motion. The methods including IMM-OIPMHT,IS-MCMC, IPMHT and so on are analyzed in comparison by simulations of thespaced-based infrared surveillance system, which show that IMM-OIPMHT works asthe best one.In chapter4, the studies are focused on TBD for the dim targets with low SNR.Firstly, the target dynamic and measurement model are set up for TBD. Secondly,heuristic particle filter on multiple-model (MM-PF) for TBD is proposed on theimprovement of the original single target MM-PF TBD, which is extended to thespace-based infrared surveillance system with unknown target number, launched andburned out time. The advantage of heuristic MM-PF TBD is validated over MM-PFTBD, histogram probability multi-hypothesis tracking(H-PMHT TBD), dynamicprogramming algorithm(DPA TBD) and so on. Lastly, to settle the low efficiency of theiterative searching of heuristic MM-PF, probability hypothesis density (PHD) TBDproposed recently is introduced and improved in combination with multiple-model(MM-PHD TBD) to track the dim targets with nonlinear motion. The experiments forspace-based infrared surveillance system with dim targets show that MM-PHD TBD assame as heuristic MM-PF outperforms other TBD algorithms such as PHD, H-PMHTand DPA.更多还原