【作者】 祝依龙；

【导师】 郭桂蓉； 付强；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2011， 博士

【摘要】 机动检测是属于一类典型的突变检测问题,是行为辨识、目标跟踪与识别等研究领域所广泛关注一个重要课题,在机动目标跟踪与拦截、自动驾驶、汽车防撞等军、民用领域都有着广泛的应用。然而目前的机动检测技术主要依赖滤波过程的观测残差及其统计特征,难以同时获得良好的机动检测概率和快速的机动检测性能。由于雷达目标特征信号与其运动模式之间具有紧密的关系,因而雷达回波中蕴含了丰富的目标运动模式相关信息,从而为利用特征信号进行机动检测提供了可能性。本文结合防空、反导、远程空空等应用背景,以脉冲多普勒雷达系统为主要对象,通过研究雷达目标特征信号,旨在设计低检测延迟的机动检测器以提高检测概率,开展了“特征辅助的目标机动检测技术”研究。第一章在分析机动检测技术存在的问题后,指出本文的研究思路。首先结合应用背景分析特征辅助的机动检测技术的重要意义;然后对机动检测问题的国内外研究现状进行较为全面的总结和分析;在此基础上指出目前存在的不足和需要进一步研究的方向和内容,最后介绍论文的主要工作和安排。第二章通过分析机动目标雷达特征信号,建立雷达特征信号与目标运动模式之间的联系,为特征辅助的机动检测技术提供理论依据。首先推导分析不同运动模式下目标姿态角变化率公式,结果表明机动目标跟踪与机动检测领域中重点关注的法向加速机动与非机动模式具有很好的可分性;其次基于多散射中心目标模型,推导姿态敏感雷达特征信号,得到雷达散射截面(Radar Cross Section, RCS)、角闪烁线偏差在不同收发极化方式下矢量形式表达式,给出了目标高分辨多普勒像的数学表示;最后以两理想散射点组成的简单目标为例,分析目标特征信号变化与姿态角变化率之间的关系,得到不同运动模式下雷达特征信号的起伏变化特性,从而建立了雷达特征信号与目标运动模式之间的联系。第三章应用特征分析的结论,设计了一种新的基于多普勒像的机动检测器,并给出新的性能评估指标。首先介绍了已有的基于RCS和角闪烁特征信号的机动检测算法,并在高数据率应用条件下对算法进行了改进,提出基于累积检验统计量的机动检测器;然后介绍了高分辨多普勒像的性质以及成像处理流程,并基于多普勒像特征信号提出一种新的机动检测方法,通过提取多普勒像分辨率差异特征矢量,将目标机动检测问题视作非机动、机动两类分类识别问题,构造一种基于后向传播神经网络的模式分类机动检测器;随后提出两个新的检测延迟时间评估指标,即延迟常数和收敛时间,相对于平均检测延迟,这两个新的指标体现了对应时刻检测器的检测概率,更全面地反映检测器动态性能;最后通过仿真实验全面评估特征辅助机动检测算法的性能,实验结果表明特征辅助机动检测算法有效降低了机动检测延迟,大大提高了机动检测概率,全面提升了机动检测性能,并且本文提出的基于多普勒像的机动检测算法性能总体上优于其它两种特征辅助的机动检测算法。第四章为实验验证部分,分别通过外场和仿真实验验证了基于雷达特征信号的运动模式的可分性以及特征辅助的机动检测器的有效性。首先针对外场实验中雷达中频信号采集问题,讨论了相参脉冲雷达中频信号采集系统的设计,重点分析同步触发系统的通用性设计,包括采样率约束条件和脉间相参条件不满足时非均匀采样信号重建问题等,并基于虚拟仪器技术构建了高速大容量采集记录系统,成功应用于本次外场实验的信号采集任务;其次介绍了外场实验系统组成情况,给出了外场实验场景设置,由于受限于多种因素,未能直接验证机动起始和机动终止检测过程,但实验数据处理结果表明,基于RCS和多普勒像特征信号构造的检验统计量可有效区分非机动和机动运动模式,为特征辅助的雷达目标机动检测算法的工程应用进行了有益的探索;最后对特征辅助的机动检测器在机动目标跟踪中的有效性进行了验证,分析了机动检测延迟与状态估计误差之间关系,给出了在一定的误差上限条件下的机动检测延迟上限,为机动检测器设计提供了理论依据,并提出了特征辅助的变结构多模机动目标跟踪算法,仿真实验结果表明所提出的算法性能优于自主多模和交互多模算法,并且计算量更小。第五章总结了本文的研究工作及所取得的结论和成果,并指出了论文研究中存在的问题和需要进一步研究的方向。更多还原

【Abstract】 Maneuverdetection belongs to the classic problem of abrupt change detection. It hasattracted considerable attention in activity identification, target tracking and recognition,and other communities, and has a wide range of military and civil applications, e.g., ma-neuvering target tracking and interception, automatic drive, car anti-crash, etc. However,at present, maneuver detection technology mostly depends on measurement residual anditsstatisticsinthefiltering. Itcannotobtainhighdetectionprobabilityandquickdetectionspeed simultaneously. Due to the close relationship between radar signatures and targetmotion mode, rich information of motion modes is embedded in the echoes. This pro-vides the feasibility of maneuver detection using radar signatures. Combining with aerialdefence, antimissile defence, long range air-to-air defence, and other application back-grounds, this thesis employs pulse Doppler radar system as the main object, and aimsto design a maneuver detector with low detection delay and high detection probabilityby studying radar signatures, which is entitled“feature aided target maneuver detectiontechnology”.Chapter 1 analyzes the existing problems in the maneuver detection, and presentsthe route of our research work. We first analyze the major significance of feature aidedmaneuver detection combing with application backgrounds. The researching progress ofmaneuver detection at home and abroad is summarized and surveyed comprehensively.We then point out current shortcomings and further researching directions and work. Ourwork and organization of the thesis are introduced at last.Chapter 2 establishes the relationship between radar signatures and target motionmodes by analyzing the signatures of maneuvering targets. It provides theoretical guide-lines for feature aided maneuver detection. Firstly, the formulae of target pose angularrates under different motion modes are derived. The results show that normal acceleratedmaneuvering motion mode and nonmaneuvering motion mode are well classifiable. Theformer has attracted much attention by maneuvering target tracking and maneuver detec-tion. Secondly, the expressions of pose sensitive radar signatures including radar crosssections (RCS) and angular glint errors under different polarizations are obtained in thevector form based on a multiple-scatterer model. Mathematical description of high reso-lution Doppler profile (HRDP) is also given. Finally, taken a simple target consisting of two ideal scatterers as an example, we analyze the relationship between radar signaturesand pose angular rates, and come to the conclusions on the fluctuating characteristics ofthe radar signatures.Chapter 3 designs a novel maneuver detector based on the HRDP by applying theconclusions of signature analysis, and presents novel indices of performance evaluation.Firstly,existingalgorithmsareintroducedwhicharebasedonRCSandangularglinterror.We then modified the algorithms to accommodate to high data rate applications, and pro-posethenewdetectorsbasedonthecumulativeteststatistics. Secondly,thepropertiesandprofiling scheme of HRDP are introduced. A novel maneuver detection algorithm is pro-posed based on the HRDP. We extract the feature vector of HRDP resolution differences,regard maneuver detection as classification problem where nonmaneuvering and maneu-vering motion modes are two classes to be classified, and develop a maneuver detectorbased on the back propagation neural network. Thirdly, two indices, i.e., delay constantand convergence time, are proposed to evaluate the detection delay performance. Theyreflect the detection probabilities at the corresponding times with respect to the averagedetection delay, and reflect the dynamic performance of the detector more completely.Finally, the performance of feature aided maneuver detection algorithms is fully evaluat-ed by simulation experiments. The results show that the feature aided maneuver detectionalgorithms decrease maneuver detection delay effectively, improve detection probabilitygreatly, and enhance detection performance generally. In addition, the proposed maneu-ver detection algorithm which is based on HRDP outperforms the other two feature aidedalgorithms.Chapter 4 is a chapter of the experimental validation. The field and simulated exper-iments are carried out to validate the classifiability of motion modes based on radar signa-tures and the effectiveness of feature aided maneuver detectors. Firstly, aiming at radar IFsignal smapling problem in the field experiments, we discuss IF signal sampling systemdesign for coherent pulse radar. The generality design of synchronous trigger system ispresented, which includes constraints on sampling rate, nonuniformly sampled signal re-construction when inter-pulse coherent conditions are not met. A sampling and recordingsystem with high speed and huge capacity is developed based on the virtual instrumen-t technique, and is applied to the signal sampling tasks in the current field experimentsuccessfully. Secondly, the components of the experimental system are introduced, and the experimental scenarios are presented. Maneuver onset and termination detection isnot validated directly due to the limitations of many factors. Fortunately, experimentalresults show that the test statistics based on RCS and HRDP can distinct nonmaneuveringand maneuvering motion modes. It is a valuable exploration for feature aided target ma-neuver detection algorithms in engineering application. Finally, the effectiveness of thefeature aided maneuver detector in the application of maneuvering target tracking is vali-dated. The filtering error dynamics in terms of detection delay are presented and a upperbound for detection delay with given filtering errors is given, which provide theoreticalguidelinesformaneuverdetectordesign. Afeatureaidedvariable-structuremultiplemod-el approach for maneuvering target tracking is proposed. Simulation results show that theproposed algorithm outperforms the autonomous multiple model and interacting multiplemodel algorithms, and provides less computational complexity.Chapter 5 summarizes the researching work of the thesis, and presents the resultsand conclusions we obtained. The shortcomings and future work are also included.更多还原