【作者】 张瑞；

【导师】 黎湘；

【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】 高分辨距离像（HRRP）以易于获取和处理、工程代价合理等优势而在雷达目标识别中占据重要地位。受应用场合和电磁散射计算技术发展的限制，基于现有目标建模和识别算法的HRRP识别技术难以取得令人满意的实用性能。以战场侦察和态势感知为需求背景，本文在HRRP姿态敏感性机理分析和识别方法研究两大方面展开工作，为雷达目标HRRP识别技术的实用化提供技术支撑。第一章阐述了论文的研究背景和意义，回顾了雷达自动目标识别技术的发展，总结了姿态敏感性、平移敏感性、预处理、特征提取和分类器设计等关键技术的研究现状，指出了现有HRRP识别技术手段存在的问题。第二章介绍HRRP的基本理论。基于理想散射点模型假设，阐述了HRRP的形成原理和数学模型，重点分析了HRRP的姿态敏感性。基于HRRP的数学模型，推导了HRRP对理想点目标的分辨能力，完善了雷达功能分辨力的定义，揭示了功能分辨力的产生原因——雷达有限带宽导致理想散射点在HRRP上体现为主瓣具有一定宽度的sinc函数，引出了HRRP的闪烁现象，为下一章的工作奠定基础。第三章研究HRRP的姿态敏感性机理。（1）发现了HRRP的闪烁现象及其“伪双峰”特点，建立了闪烁HRRP的数学模型，推导了闪烁现象的精确发生条件，得到了闪烁现象发生概率与雷达、目标参数的关系，并用暗室数据验证了上述理论分析结果。（2）推导了闪烁HRRP与平均HRRP的匹配度特性，以此为依据实验研究了闪烁现象对固定角域划分和基于相关系数的自适应角域划分模板数量的影响，为HRRP模板库的建立提供理论基础。（3）定性分析了闪烁对强散射中心、径向长度等HRRP结构特征的影响，并基于匹配度这一数学工具度量了闪烁对幅度谱、功率谱等HRRP谱特征的影响，为HRRP特征提取提供指导。第四章研究HRRP姿态敏感性的利用问题。基于理想散射点模型假设下的高分辨回波模型，推导得到了距离单元回波功率随慢时间的近似余弦变化规律，发现了回波功率-慢时间频谱中蕴含的目标“局部”横向尺寸信息。结合HRRP的径向分辨特性，提出了基于HRRP幅度起伏特性的目标横向尺寸快速估计方法。相比传统基于ISAR像的横向尺寸估计方法，所提方法避免了ISAR成像的繁琐步骤且横向上只需一次傅里叶变换，因此运算量和估计误差大幅度降低。仿真和暗室数据的实验结果证明了该方法的有效性和低复杂度。第五章研究HRRP的幂变换预处理技术。利用Gaussian分布的统计特性，提出了基于松弛偏峰度正态检验的幂变换参数学习方法。在此基础上，分析了三种典型距离单元回波的幂变换特点，并结合HRRP的高维特性和姿态敏感性，提出了HRRP的自适应幂变换预处理方法。该方法为不同目标、不同角域的HRRP以及HRRP的不同距离单元自适应选择幂变换参数，能够显著提升HRRP样本的Gaussian性能，充分发挥线性分类器的潜力。基于MSTAR实测数据的实验证明了该算法的有效性。第六章研究HRRP的特征提取技术。提出了基于距离单元间稳定性差异的加权HRRP特征，该特征具有较松弛的方位敏感性。在此基础上，结合HRRP样本在单位超球面上的“局部紧凑+整体稀疏”分布特点，建立了“类心+紧密度球”特征模板库，定义了基于紧密度球的距离度量准则，提出了基于样本紧密度的HRRP识别方法。该方法充分利用了HRRP样本的空间分布特性，提高了HRRP的识别性能。基于MSTAR实测数据的实验证明了该算法的有效性。第七章研究HRRP的分类器设计。基于HRRP的稀疏特性和姿态敏感性，定义了散射中心模型表示残差，设计了散射中心模型-稀疏表示分类器（SCM-SRC）。SCM-SRC能够避免传统SRC应用于HRRP识别时的散射中心模型间的“过拟合”问题，提高识别性能。基于MSTAR实测数据的实验证明了该算法的有效性。论文最后总结了全文工作，并指出了下一步要着力开展的研究内容。更多还原

【Abstract】 High resolution range profile (HRRP) plays an important part in the radarautomatic target recognition community for the simplicity in acquiring and processing,the feasibility in engineering, and so on. However, the HRRP recognition performancebased on the popular template database and recognition algorithms is not alwayssatisfying for some factors, such as the constraint at the application background and thedevelopment in electromagnetic calculation. In allusion to the requirement in battlefieldreconnaissance and situation awareness, this dissertation focuses on the research on theazimuth sensitivity and recognition algorithms of HRRP and provides some technicalsupports for the engineering application of HRRP recognition algorithms.Chapter1presents the background of this dissertation. We review the developmentof radar automatic target recognition and summarize some key techniques in HRRPrecognition, which are the azimuth sensitivity, shift sensitivity, preprocessing, featureextraction and classifier selection. We also point out the disadvantage of abovetechniques.Chapter2introduces the basic theory of HRRP. Based on the scattering centremodel, we investigate the acquiring processing and mathematical model. Then, wemake profound analysis about the azimuth sensitivity. Based on the mathematical modelof HRRP, we conclude the resolution ability of HRRP for two ideal point targets andconsummate the definition of functional resolution. The functional resolution isessentially induced by the limit bandwidth of radar transmit signal, which makes theideal point target spread as the sinc function in HRRP. Also, we demonstrate the speckleof HRRP and provide the theoretical basis for the work in next chapter.Chapter3focuses on the azimuth sensitivity of HRRP. Firstly, we announce thespeckle of HRRP and the “spurious dual peaks” feature of speckled HRRP. Weestablish the theoretical model of speckle and conclude the occurrence condition ofspeckle. Thereby, we obtain the relationship between the speckle probability in HRRPand the parameters of radar and the target. The experiment in an anechoic chamber isused to verify all the analyses about the speckle. Secondly, we conclude the matchingscore between speckled HRRP and average HRRP. Based on the conclusion, we studyexperimentally the influence of the speckle on the number of average HRRPs, which arerespectively obtained according to the constant angular production and the adaptiveangular production. The obtained results can provide some theoretical supports for thetemplate database forming. Thirdly, we do some research on the influences of thespeckle on the structural features, such as the predominant scatterers and the rangelength. And, the influences of the speckle on some spectral features are measured by thematching score. These works are meaningful for the selection of HRRP feature. Chapter4studies the application of the azimuth sensitivity of HRRP. Based on themathematical model of high resolution radar echo, we demonstrate that the fluctuationof echo power with the low time can be deemes as the summation of several sincfunctions. And, the “partial” cross-range length of the target can be extracted from thespectrum of the curve of echo power-low time. Then, making use of the range resolutionof HRRP, we propose a fast algorithm for estimating the cross-range length of radartarget using the fluctuation of HRRP. Compared with the conventional method based onISAR imaging, the proposed algorithm avoids some complicate steps and takes only oneFast Fourier Transform at the cross-range. The experimental results demonstrate theeffectiveness of the proposed algorithm and its low computational complexity.Chapter5studies the BOX-COX transformation of HRRP. Based on the statisticalcharacteristic of Gaussian distribution, we demonstrate a method to estimate theBOX-COX transformation parameter using the skewness and kurtosis normal test. Then,we make a deep insight into the BOX-COX transformation characteristics of three typesof typical echo. Considering with the high vector nature and azimuth sensitivity ofHRRP, we propose a new algorithm to perform the BOX-COX transformation forHRRP. The propose algorithm adaptively adjusts the parameter of BOX-COXtransformation for each angular frame of each target and each range cell of HRRP.Therefore, it can improve the normality performance of HRRP and inspire the ability ofthe classifier with the linear discriminant function. Experimental results for MSTARdata show that the proposed algorithm can improve the recognition performancesignificantly.Chapter6studies the feature extraction of HRRP. Based on the difference betweenrange cells, we propose a new weighted HRRP which is more immune from the azimuthof the target. Considering that HRRPs are located on the unit hypersphere withclustering only in some areas, we establish the template of center-affinity sphere. Then,we define a new distance according to the affinity sphere and propose an algorithmbased on the affinity among HRRP samples. The proposed algorithm utilizes the spatialcharacteristic of HRRPs and can improve the recognition performance significantly.Experimental results for MSTAR data show the validity of the proposed algorithm.Chapter7studies the classifier of HRRP. Based on the sparse characteristics andazimuth sensitivity of HRRP, we define the residual of the scattering center model.Then, we design the classifier called the scattering centre model and sparserepresentation-based classifier (SCM-SRC). SCM-SRC can avoid the overfittingphenomenon when the typical SRC are applied to HRRP recognition. Experimentalresults for MSTAR data show the validity of the proposed algorithm.Chapter8summarizes the dissertation and discusses the future work to beresearched.更多还原