基于高分辨距离像的雷达目标识别研究

发布时间：2018-02-25 22:06

本文关键词： 雷达目标识别高分辨距离像流形学习几何结构特征谱包络信息融合　出处：《电子科技大学》2016年博士论文　论文类型：学位论文

【摘要】：探测与测距是早期雷达的基本功能,这已经远远不能满足现代雷达需要获取越来越多的目标信息的需求。在军用和民用的很多应用中,不但需要探测到目标,还要识别出是什么目标,即雷达目标识别。目标识别自然成为现代雷达信息处理中非常重要的研究方向之一。雷达信号带宽的提高使得雷达具有距离向高分辨能力,可对目标进行高分辨成像。高分辨距离像(HRRP)能够较好的表征观测目标等效多散射中心沿距离向的分布结构,且易于获取和处理,为我们提供了一种非常有潜力的雷达目标识别手段。以高分辨距离像为研究对象,围绕着稳健特征提取、多特征综合、多特征信息融合、系统构架等关键问题,对雷达目标高分辨距离像识别中所涉及的相关理论和关键技术开展了深入的理论研究和实验验证。论文主要工作和创新之处概况如下:(1)对两种典型的流形学习算法——邻域保持投影(NPP)和局部切空间排列(LTSA)进行研究,分析了算法具备松弛HRRP的姿态敏感性的优良特性。针对HRRP雷达目标识别,分别提出了增强的邻域保持投影(ENPP)算法和增强核邻域保持投影(EKNPP)算法,以及线性鉴别局部切空间排列(LDLTSA)算法和核鉴别局部切空间排列(KDLTSA)算法。实验结果验证了所提算法的有效性以及相较于现有的同类算法所表现出来的性能优势。(2)针对雷达HRRP目标识别中由于训练样本非常有限导致传统的子空间算法学习性能下降的问题,对基于点到空间距离测度的子空间学习算法进行分析和研究,提出了两种新的基于点到空间距离测度的学习算法:邻域特征空间鉴别分析I(NFSDA-I)和邻域特征空间鉴别分析II(NFSDA-II)。实验结果表明,相对于其它已有的点到空间类的学习算法,NFSDA-I和NFSDA-II算法的子空间具有更高的多目标鉴别能力,目标识别性能较优。(3)对HRRP时域回波中潜在的目标几何结构特征进行分析,采用统计的方法,从HRRP时域回波中提取出8个从不同角度反映目标几何结构信息的特征量,并采用多特征综合的研究思路,选择多个特征组合起来得到8个综合特征。实验结果表明了其中一些几何结构特征的有效性,如:熵和不规则度特征,以及多特征综合识别所具有的性能优势。(4)首次将语音识别领域里有关谱包络的研究成果引入到HRRP雷达目标识别中,从HRRP的频域特性中提取出9个典型的谱包络特征,并组合构建了21个综合特征,用于目标分类。实验结果表明,所提取的谱包络特征对于HRRP雷达目标识别是有效的,且具有一定的潜力。此外,采用多个谱包络特征综合识别的效果良好。(5)研究了基于多特征融合的雷达目标识别技术。对基于信息融合的雷达目标识别系统框架和相关的融合算法进行了研究,在此基础上,给出了一个基于Dempster-Shafer理论多特征融合的HRRP雷达目标识别方案,分别提取四种不同特征、采用两种分类器进行分类,并在决策层上基于Dempster-Shafer理论进行融合判决。实验表明了该融合识别方案的有效性。(6)对宽带数字阵列雷达目标识别系统进行研究。以S波段16阵元线阵的宽带数字阵雷达系统为基础,构建了基于OpenVPX的信号与信息处理系统,并建立了适用于串行高速总线的目标识别开放式软件架构。
[Abstract]:Is the basic function of early detection and ranging radar, which can not meet the needs of modern radar to get more information of the target in the military and civilian needs. In many applications, not only need to detect the target, but also identify what goal, namely the radar target recognition. Target recognition has become one of very important research direction of information in the processing of modern radar. The radar signal because of the increasing bandwidth of radar has high range resolution ability to target high resolution imaging. High resolution range profile (HRRP) to the distribution characterization of the target equivalent better scattering center along the range direction, and is easy to acquire and process, provides a very the potential of the radar target recognition method for us in HRRP as the research object, around the robust feature extraction, multi feature and multi feature information fusion system. The key problem of structure, on the radar target HRRP recognition in the related theory and key technology to carry out in-depth theoretical and experimental research. The main work and innovations of the paper are as follows: (1) survey of two typical manifold learning algorithm, neighborhood preserving projection (NPP) and local tangent space alignment (LTSA) research, analysis of the excellent properties of sensitivity of HRRP relaxation algorithm with the attitude. The HRRP radar target recognition, are put forward to enhance the neighborhood preserving projection (ENPP) algorithm and the enhanced kernel neighborhood preserving projection (EKNPP) algorithm, and the identification of linear local tangent space alignment (LDLTSA) algorithm and kernel discriminant local tangent space alignment (KDLTSA) algorithm. The experimental results verify the performance advantage of the effectiveness of the proposed algorithm and compared with the existing similar algorithms is shown. (2) for radar target recognition by HRRP In the training sample is very limited in the traditional learning subspace algorithm for the problem of declining performance, based on the point to the spatial distance measure subspace learning algorithm analysis and research, put forward two new points to the space based on the distance measure learning algorithm: the neighborhood feature space I discriminant analysis (NFSDA-I) feature space and neighborhood identification analysis of II (NFSDA-II). The experimental results show that compared with other existing learning algorithms to space, NFSDA-I and NFSDA-II algorithm with subspace multi target identification ability is higher, the performance of target recognition is better. (3) the target geometry features of potential HRRP echo was analyzed by using the statistical methods to extract 8 reflect the target geometry information from different angles of the features from the HRRP echo, and the research ideas of integrated multi features, select multiple features combination up To the 8 features. The experimental results show that the effectiveness of some geometric features such as entropy and irregular features, and comprehensive recognition of the characteristics of performance advantages. (4) for the first time on the spectral envelope research results in the field of speech recognition is introduced to the HRRP radar target recognition, from the frequency characteristics of HRRP extracted from 9 typical spectral envelope features, and established 21 comprehensive features, used for target classification. The experimental results show that the spectral envelope of the extracted features are effective for HRRP radar target recognition, and has a certain potential. In addition, the multi spectral envelope feature recognition the effect is good. (5) studied the technology of radar target recognition based on multi feature fusion of radar target recognition system framework and related fusion algorithm based on information fusion is studied, on this basis, based on a given De The theory of mpster-Shafer multi feature fusion HRRP radar target recognition scheme, were extracted from four different characteristics, using two kinds of classifier and decision level fusion based on the Dempster-Shafer theory of judgment. Experimental results show the validity of the fusion scheme. (6) research on wideband digital array radar target recognition system. S band 16 element linear array of wideband digital array radar system based on OpenVPX is constructed based on signal and information processing system, and the establishment of a suitable for high-speed serial bus target recognition of open software architecture.

【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN957.51

【参考文献】