【作者】 侯庆禹；

【导师】 保铮；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2009， 博士

【摘要】 雷达目标的高分辨距离像(HRRP)可以反映目标散射点沿距离方向的分布情况,提供了目标重要的结构信息,被广泛用作雷达目标的分类与识别,成为雷达自动目标识别研究领域的一个热点。同时,随着军事需求的日益迫切,需要研究人员加快对目标识别的理论研究向工程实现迈进。因此,本论文主要围绕着国防预研及国家自然科学基金的相关项目,针对雷达高分辨距离像目标识别,从在杂波和噪声环境下的HRRP稳健性识别、基于变分贝叶斯(VB)方法的分类器设计以及对HRRP的特征提取与分层目标识别等三个方面进行了的研究。本论文的主要内容可概括为如下六个部分:第一部分,简要给出了雷达自动目标识别的基本概念,并列举了国内外基于HRRP自动目标识别的研究进展,介绍了本文的研究工作。第二部分,研究了HRRP在杂波情况下如何保持稳健性识别性能的问题,关键技术是如何抑制杂波。与以检测为目的的杂波抑制相比,宽带目标识别雷达的杂波抑制需要在抑制杂波的同时,尽可能地保持目标信号的结构信息不变,这样才能进行下一步的目标识别任务。为达到这一目的,我们先后提出了三种不同的宽带目标识别雷达的杂波抑制方法。(1)设计滤波器直接在多普勒域将杂波滤除。该算法主要是利用杂波的起伏速度通常不是很大,在脉间的相关性比较强,因此可以通过在多普勒域杂波抑制后,再将信号变换回时域,进行相干积累,提高目标的信噪比。(2)在宽带雷达下,目标速度较大时,容易出现越距离单元走动(MTRC)的现象,方法1其实是没有考虑这一点的,因此我们进而提出先利用keystone变换校正目标MTRC,然后再对其采用方法1来抑制杂波,另外针对目标出现多普勒模糊而杂波没有模糊时,采用在频率-多普勒域直接将目标信号部分提取出来的方法来降低杂波的影响。(3)针对目标出现MTRC的现象,提出了另外的解决思路,不需要通过keystone变换校正MTRC,在频率-多普勒域利用Hough变换将目标信号部分的线段提取出来,同时如果可以大致估计出目标的速度,还可以采用更加简单的方法来处理,即将目标做运动补偿后,在频率-多普勒域提取对应的目标信号部分。第三部分,研究了在噪声背景下的HRRP稳健识别问题。当目标与雷达距离较远时,其信噪比将会降低,因此识别算法对噪声的稳健性是HRRP目标识别在实际应用中需要解决的一个问题。我们基于概率主分量分析(PPCA)和自适应高斯分类器(AGC)模型分别提出了两种不同的算法,使得被噪声污染的测试样本能够较好地匹配在弱噪声样本条件下训练出来的模板。第四部分,研究了将变分贝叶斯(VB)算法结合目前常用的一些统计模型来解决雷达HRRP目标识别问题。VB方法为近些年被广泛用于近似求解Bayes积分的方法,通过将Bayes积分表达式中所有参数和隐变量的联合概率分布简化为各个参数以及隐变量之间概率分布的乘积,即假设各参数以及隐变量是相互独立的,这样积分表达式的值就可以利用简单的形式来代替其下界,通过不断优化参数的值来提高其下界,使得下界不断逼近该积分表达式的真实值。基于VB方法,我们将Gaussian Mixture模型和混合因子分析(FA)模型应用到雷达HRRP目标识别中来,取得了不错的效果。第五部分,利用对HRRP提取的新特征,对分层雷达目标识别做了相关研究。由于雷达HRRP可以反映目标散射点沿距离方向的分布情况,因此我们提取了关于目标尺寸大小的结构特征,即目标的支撑区长度。利用该特征首先对目标的大小做出初步分类,然后再利用常规目标识别算法进行更加精确的型号识别。同时针对螺旋桨飞机相邻回波的能量变化要大于喷气式飞机这一特点,我们提取了相邻回波之间相对能量差值大小这一特征来初步区分这两类飞机。第六部分,对全文工作进行了总结,并对下一步需要研究的工作提出了建议。更多还原

【Abstract】 Target high-resolution range profile (HRRP) represents the projection of the complex returned echoes from the target scattering centers onto the radar line-of-sight (LOS). It contains the target structure signatures, such as target size, scatterer distribution, etc., and thereby radar HRRP target recognition has received intensive attention from the radar automatic target recognition (RATR) community. Due to the increasing military demand, RATR are required to stride to practical realization from theoretical. In this dissertation, the theory and techniques for radar HRRP target recognition are researched from the three aspects, i.e. robust of HRRP recognition performance under the clutter and noise, the classifier designing based on variational Bayesian (VB), feature extraction and layered algorithm for target recognition, which are supported by Advanced Defense Research Programs of China and National Science Foundation of China.The main content of this dissertation is summarized as follows:The first part begins with a brief introduction of the fundamental theories of RATR and reviews some related work of other institutes. Then the main work of this thesis is introduced.The second part focuses on the robustness of HRRP recognition performance under the clutter environment. The key point is how to suppress clutter. Compared with the clutter suppression for target detection, clutter suppression for wideband target recognition radar requires that the target structure signatures are not changed after the clutter is suppressed. We present three methods of clutter suppression for wideband target recognition radar to achieve this purpose. (1) Clutter is suppressed by a filter in Doppler domain. This algorithm mainly exploits the fact that the velocity of clutter is small, and the correlation of clutter between different pulses is high. After the clutter suppression, we can transform the signal to the time domain, and then perform coherent accumulation, aiming at improving the signal-to-noise ratio (SNR) by. (2) In wideband radar, the target’s migration though resolution cells (MTRCs) will occur when the velocity is high. But MTRC is not considered in Algorithm 1. Thereby, we utilize keystone formatting to mitigate the MTRCs, and then suppress clutter by algorithm1. Otherwise, if target has Doppler ambiguity while clutter does not, we extract the target directly to reduce the effect of clutter in Doppler-frequency domain. (3) If MTRC occurs, we can utilize the Hough transform to extract the line segment of target in Doppler-frequency domain even though MTRC is not mitigated via keystone formatting. A simple method is proposed if the velocity of target can be approximately estimated, which is extracting the line segment of target in Doppler-frequency domain after motion compensation.The third part is contributed to noise robust in HRRP target recognition. The SNR will be decreased when target is far away from radar, and therefore, the robustness study of HRRP recognition algorithm is necessary. In this part, based on PPCA model and AGC model, a robust algorithm for HRRP statistical recognition is presented when test SNR is lower than training SNR.The fourth part focuses on radar HRRP statistical recognition based on VB. VB method is widely used to approximately resolve Bayesian integral in recent decade. On the assumption that parameters and hidden variables are independent of each other, the jointly probability distribution over all parameters and hidden variables can be approximated with a simpler distribution which is a lower bound of original Bayesian integral. The lower bound is increased by optimizing parameters, and the aim is to approximate the real value of original Bayesian integral. We apply Gaussian mixture models and mixtures of factor analyzers model to radar HRRP statistical recognition based on VB method, and obtain a good performance with measured radar data.In the fifth part, utilizing the new feature extracting from HRRP, the layered radar target recognition is focused. Due to the fact that HRRP represents the projection of the complex returned echoes from the target scattering centers onto the radar line-of-sight (LOS), we extract target size, one of the target structure signatures. First, we utilize this feature to classify different targets by their size, and then identify them exactly by normal target recognition algorithm. In addition, we can distinguish propeller-driven aircraft from jet plane by relative difference energy between coherent echoes, because the relative difference energy between coherent echoes of propeller-driven aircraft is large than jet plane’s.Finally, we summarize the main results of the study which have led to this thesis; additionally, some conclusions are drawn and some recommendations for future work are given.更多还原