基于一维距离像的目标检测与鉴别方法
发布时间：2018-09-07 11:06
【摘要】：随着距离分辨率的提高,目标能量分布在雷达回波中的多个距离单元内,被称作距离扩展目标或者分布式目标。距离扩展目标的回波之中包含目标更多的信息,如何有效地利用这些信息成为雷达技术领域迫切需要解决的问题。目前针对雷达高分辨距离像的目标识别问题已经得到了广泛的关注,但对于距离扩展目标的检测更多地停留在理论研究层面,如何对检测结果进行鉴别,有效地去除检测结果中虚警的研究工作则很少涉及。本文对距离扩展目标的检测方法进行研究,并将目标的检测结果和一类分类器相结合,提出了一种基于一维距离像的目标鉴别方法。本文的主要工作如下:1.对线性调频步进信号和雷达杂波统计模型进行研究。给出了线性调频步进信号的参数选取准则、速度补偿方法和高分辨距离像合成方法,并结合仿真结果进行比较。然后介绍了雷达杂波数据的功率谱模型和幅度的统计模型,给出了两种比较常用的杂波模拟方法:零记忆非线性变化法和球不变随机过程法。2.对距离扩展目标的检测问题进行研究。介绍了统计信号的检测理论,给出了奈曼-皮尔逊准则和恒虚警率检测的含义。接着给出了几种针对点目标的检测方法和对应的检测门限的计算方法,并结合仿真结果说明计算得到的检测门限可以保证虚警概率的近似不变。然后介绍了基于二进制积累的检测方法,提出了瑞利杂波模型和对数正态杂波模型下距离扩展目标的直接积累检测法,并给出相应的检测统计量与检测门限的计算方法。最后通过仿真实验对几种检测方法进行比较。3.针对一维距离像的目标鉴别问题进行研究。首先,结合实例对最近邻一类分类器、K-近邻一类分类器以及K-中心一类分类器的主要思想和分类过程进行说明。接着引入衡量两个点集间差异性大小的Hausdorff距离,对最近邻一类分类器进行改进,结合目标检测的结果提出了一种针对一维距离像的目标鉴别方法,并将该鉴别方法推广到K-中心一类分类器。然后基于实测数据,详细分析了一类分类器的不同参数对鉴别性能的影响,并比较了欧氏距离和Hausdorff距离的鉴别性能,验证了提出的鉴别方法的有效性,结合实验比较了改进的最近邻一类分类器和K-中心一类分类器在鉴别过程中各自的优势。
[Abstract]:With the improvement of range resolution, target energy is distributed in multiple range units in radar echo, which is called range extended target or distributed target. The echo of the extended range target contains more information of the target. How to utilize this information effectively has become an urgent problem in the field of radar technology. At present, the problem of target recognition of radar high resolution range profile has been paid more attention to, but the detection of extended range target is more focused on the theoretical research, how to identify the detection results. The research work of removing false alarm effectively is seldom involved. In this paper, the detection method of extended range target is studied, and a target discrimination method based on one-dimensional range profile is proposed by combining the detection results of target with a class of classifiers. The main work of this paper is as follows: 1. The statistical model of linear frequency modulation step signal and radar clutter is studied. The parameter selection criterion, velocity compensation method and high resolution range profile synthesis method of LFM stepper signal are presented, and the simulation results are compared. Then, the power spectrum model and amplitude statistical model of radar clutter data are introduced, and two common clutter simulation methods are given: zero memory nonlinear variation method and spherical invariant random process method. The detection of extended range targets is studied. This paper introduces the detection theory of statistical signal, and gives the meaning of Neiman-Pearson criterion and CFAR detection. Then several detection methods for point targets and the corresponding detection threshold calculation methods are given, and the simulation results show that the calculated detection threshold can guarantee the approximate invariance of false alarm probability. Then the detection method based on binary accumulation is introduced, and the direct accumulation detection method for extended distance target under Rayleigh clutter model and logarithmic normal clutter model is proposed, and the corresponding detection statistics and detection threshold are calculated. Finally, several detection methods are compared by simulation experiments. 3. 3. The target identification problem of one-dimensional range profile is studied. Firstly, the main ideas and classification processes of the nearest neighbor class classifier and the K- center class classifier are explained by an example. Then the Hausdorff distance is introduced to measure the difference between the two sets of points, and the nearest neighbor classifier is improved, and a target discriminating method for one dimensional range profile is proposed based on the result of target detection. The discriminant method is extended to K-center classifier. Then, based on the measured data, the effects of different parameters of a class of classifiers on the discriminant performance are analyzed in detail, and Euclidean distance and Hausdorff distance are compared to verify the effectiveness of the proposed discriminant method. The advantages of the improved nearest neighbor classifier and the K- center classifier in the discriminant process are compared with experiments.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
本文编号：2228083
[Abstract]:With the improvement of range resolution, target energy is distributed in multiple range units in radar echo, which is called range extended target or distributed target. The echo of the extended range target contains more information of the target. How to utilize this information effectively has become an urgent problem in the field of radar technology. At present, the problem of target recognition of radar high resolution range profile has been paid more attention to, but the detection of extended range target is more focused on the theoretical research, how to identify the detection results. The research work of removing false alarm effectively is seldom involved. In this paper, the detection method of extended range target is studied, and a target discrimination method based on one-dimensional range profile is proposed by combining the detection results of target with a class of classifiers. The main work of this paper is as follows: 1. The statistical model of linear frequency modulation step signal and radar clutter is studied. The parameter selection criterion, velocity compensation method and high resolution range profile synthesis method of LFM stepper signal are presented, and the simulation results are compared. Then, the power spectrum model and amplitude statistical model of radar clutter data are introduced, and two common clutter simulation methods are given: zero memory nonlinear variation method and spherical invariant random process method. The detection of extended range targets is studied. This paper introduces the detection theory of statistical signal, and gives the meaning of Neiman-Pearson criterion and CFAR detection. Then several detection methods for point targets and the corresponding detection threshold calculation methods are given, and the simulation results show that the calculated detection threshold can guarantee the approximate invariance of false alarm probability. Then the detection method based on binary accumulation is introduced, and the direct accumulation detection method for extended distance target under Rayleigh clutter model and logarithmic normal clutter model is proposed, and the corresponding detection statistics and detection threshold are calculated. Finally, several detection methods are compared by simulation experiments. 3. 3. The target identification problem of one-dimensional range profile is studied. Firstly, the main ideas and classification processes of the nearest neighbor class classifier and the K- center class classifier are explained by an example. Then the Hausdorff distance is introduced to measure the difference between the two sets of points, and the nearest neighbor classifier is improved, and a target discriminating method for one dimensional range profile is proposed based on the result of target detection. The discriminant method is extended to K-center classifier. Then, based on the measured data, the effects of different parameters of a class of classifiers on the discriminant performance are analyzed in detail, and Euclidean distance and Hausdorff distance are compared to verify the effectiveness of the proposed discriminant method. The advantages of the improved nearest neighbor classifier and the K- center classifier in the discriminant process are compared with experiments.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
【参考文献】
相关期刊论文 前1条
1 吴振凯;;调频步进信号回波的速度补偿[J];制导与引信;2010年01期
，本文编号：2228083
本文链接：https://www.wllwen.com/kejilunwen/wltx/2228083.html
