基于极化信息的SAR目标检测鉴别方法研究
发布时间：2018-09-09 17:36
【摘要】：极化合成孔径雷达(Polarimetric Synthetic Aperture Radar,PolSAR)利用不同极化方式交替发射与接收雷达信号,能够获得丰富的目标散射信息。这些信息在雷达图像的特征提取,图像解译及自动目标识别(Automatic Target Recognition,ATR)技术中发挥着重要作用。因此本文从极化信息的提取与运用出发,对基于极化SAR图像的极化目标分解,基于极化信息的目标检测与目标鉴别等方面进行了研究。本文首先简要介绍了极化SAR目标分解,目标检测与鉴别的研究背景与意义,概述了该课题的国内外研究状况,并介绍了本文研究的主要内容。在此基础上,对本文的主要研究内容分以下三个方面进行详细介绍:第一部分,主要介绍了极化处理的电磁波理论基础,并在此基础上对极化数据的预处理:极化定标理论进行介绍,分别研究了基于点目标的定标算法和基于分布式目标的定标算法,分析了极化相位定标,串扰定标及通道不平衡定标对极化数据的影响。第二部分,研究了极化目标分解问题,主要是非相干分解中基于模型的分解问题。该工作主要包含以下两方面:1,介绍了几种经典的基于模型的分解算法,包括Freeman-Durden分解,改进的Yamaguichi分解,基于非负特征约束的模型分解,基于双酉变换的分解等。2,提出了一种基于极化相似度匹配的极化相干矩阵散射能量提取方法,该方法利用极化相似度获取与原始相干矩阵相似度最高的基本散射机制并对该散射机制进行半正定约束的优先分解,避免了主导散射机制能量的低估问题,取得了较好的分解结果。第三部分,研究了基于极化信息的目标检测及鉴别问题。在目标检测阶段,首先介绍了传统的基于阈值的CFAR检测算法,针对CFAR检测在复杂场景下失效的情况及对极化信息运用不充分情况,研究了基于极化信息与支持向量数据描述(Support Vector Data Discriptor,SVDD)的极化检测算法。SVDD构造了包含单类目标数据样本的紧致超球体边界,使其能很好地对一类问题进行分类。通过对目标样本提取大量极化特征,采用SVDD进行训练,获取一个较优的分类界面,采用训练得到的分类边界对待分类样本进行分类,获取目标杂波二值图,通过对二值图像进行形态学滤波去除明显不是目标的杂波区域,最终获得疑似目标切片,实现目标检测。在目标鉴别阶段,详细介绍了几种经典的鉴别特征,该特征主要是由林肯实验室及其他实验室提出的纹理特征。并介绍了高斯鉴别器及SVDD鉴别器的基本原理,采用两种鉴别器实现了目标的鉴别。
[Abstract]:Polarimetric synthetic Aperture Radar (Polarimetric Synthetic Aperture Radar,PolSAR) can obtain abundant target scattering information by alternately transmitting and receiving radar signals in different polarization modes. This information plays an important role in feature extraction, image interpretation and automatic target recognition (Automatic Target Recognition,ATR) of radar images. Therefore, from the point of view of the extraction and application of polarization information, the polarization target decomposition based on polarimetric SAR image, the target detection and target identification based on polarization information are studied in this paper. In this paper, the research background and significance of polarimetric SAR target decomposition, target detection and identification are briefly introduced, the research status of this topic at home and abroad is summarized, and the main contents of this paper are introduced. On this basis, the main contents of this paper are described in detail as follows: in the first part, the electromagnetic wave theory of polarization processing is introduced. On this basis, the preprocessing of polarization data: polarization calibration theory is introduced. The calibration algorithm based on point target and the algorithm based on distributed target are studied, and the polarization phase calibration is analyzed. The influence of crosstalk calibration and channel imbalance calibration on polarization data. In the second part, we study the problem of polarimetric target decomposition, which is mainly model-based decomposition in incoherent decomposition. This work mainly includes the following two aspects: 1, introduces several classical model-based decomposition algorithms, including Freeman-Durden decomposition, improved Yamaguichi decomposition, model decomposition based on non-negative feature constraints. Based on the decomposition of double unitary transformation, a polarization coherence matrix scattering energy extraction method based on polarization similarity matching is proposed. This method uses polarization similarity to obtain the basic scattering mechanism which has the highest similarity with the original coherent matrix, and decomposes the scattering mechanism with positive semidefinite constraints first, thus avoiding the problem of underestimating the energy of dominant scattering mechanism. A good decomposition result is obtained. In the third part, the problem of target detection and identification based on polarization information is studied. In the phase of target detection, the traditional threshold-based CFAR detection algorithm is introduced, aiming at the failure of CFAR detection in complex scenarios and the insufficient use of polarization information. In this paper, a polarization detection algorithm based on polarization information and support vector data description (Support Vector Data Discriptor,SVDD) is studied. A compact hypersphere boundary containing a single class of target data samples is constructed so that it can classify a class of problems well. By extracting a large number of polarization features from the target samples and training with SVDD, a better classification interface is obtained. The classification boundary is used to classify the classified samples, and the binary image of the target clutter is obtained. By using morphological filtering to remove the clutter region which is obviously not the target, the suspected target slice can be obtained and the target detection can be realized. In the phase of target identification, several classical discriminant features are introduced in detail, which are mainly texture features proposed by Lincoln Lab and other laboratories. The basic principles of Gao Si discriminator and SVDD discriminator are introduced.
【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：2233095
[Abstract]:Polarimetric synthetic Aperture Radar (Polarimetric Synthetic Aperture Radar,PolSAR) can obtain abundant target scattering information by alternately transmitting and receiving radar signals in different polarization modes. This information plays an important role in feature extraction, image interpretation and automatic target recognition (Automatic Target Recognition,ATR) of radar images. Therefore, from the point of view of the extraction and application of polarization information, the polarization target decomposition based on polarimetric SAR image, the target detection and target identification based on polarization information are studied in this paper. In this paper, the research background and significance of polarimetric SAR target decomposition, target detection and identification are briefly introduced, the research status of this topic at home and abroad is summarized, and the main contents of this paper are introduced. On this basis, the main contents of this paper are described in detail as follows: in the first part, the electromagnetic wave theory of polarization processing is introduced. On this basis, the preprocessing of polarization data: polarization calibration theory is introduced. The calibration algorithm based on point target and the algorithm based on distributed target are studied, and the polarization phase calibration is analyzed. The influence of crosstalk calibration and channel imbalance calibration on polarization data. In the second part, we study the problem of polarimetric target decomposition, which is mainly model-based decomposition in incoherent decomposition. This work mainly includes the following two aspects: 1, introduces several classical model-based decomposition algorithms, including Freeman-Durden decomposition, improved Yamaguichi decomposition, model decomposition based on non-negative feature constraints. Based on the decomposition of double unitary transformation, a polarization coherence matrix scattering energy extraction method based on polarization similarity matching is proposed. This method uses polarization similarity to obtain the basic scattering mechanism which has the highest similarity with the original coherent matrix, and decomposes the scattering mechanism with positive semidefinite constraints first, thus avoiding the problem of underestimating the energy of dominant scattering mechanism. A good decomposition result is obtained. In the third part, the problem of target detection and identification based on polarization information is studied. In the phase of target detection, the traditional threshold-based CFAR detection algorithm is introduced, aiming at the failure of CFAR detection in complex scenarios and the insufficient use of polarization information. In this paper, a polarization detection algorithm based on polarization information and support vector data description (Support Vector Data Discriptor,SVDD) is studied. A compact hypersphere boundary containing a single class of target data samples is constructed so that it can classify a class of problems well. By extracting a large number of polarization features from the target samples and training with SVDD, a better classification interface is obtained. The classification boundary is used to classify the classified samples, and the binary image of the target clutter is obtained. By using morphological filtering to remove the clutter region which is obviously not the target, the suspected target slice can be obtained and the target detection can be realized. In the phase of target identification, several classical discriminant features are introduced in detail, which are mainly texture features proposed by Lincoln Lab and other laboratories. The basic principles of Gao Si discriminator and SVDD discriminator are introduced.
【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
【共引文献】
中国期刊全文数据库 前5条
1 王爽;于佳平;刘坤;侯彪;焦李成;;基于双边滤波的极化SAR相干斑抑制[J];雷达学报;2014年01期
2 申晓华;王耀强;郑磊;郭小静;;Radarsat-2四极化影像在河套灌区土壤含盐量反演中的应用[J];湖北农业科学;2014年15期
3 陈建宏;赵拥军;黄洁;刘伟;赖涛;;改进的多视PolSAR非局部均值滤波算法[J];测绘科学技术学报;2014年05期
4 任俊英;苏彩霞;曹永锋;;基于中间层特征的全极化SAR监督地物分类[J];遥感技术与应用;2014年02期
5 杨靖;苏彩霞;曹永锋;;全极化SAR数据土地覆盖分类精度分析[J];遥感信息;2015年05期
中国硕士学位论文全文数据库 前10条
1 黄晓东;极化目标模型分解的不一致性研究[D];中国地质大学;2013年
2 吴祥;极化SAR与可见光遥感影像融合算法研究[D];杭州电子科技大学;2014年
3 孙晨;基于子孔径分解的极化SAR图像分类方法研究[D];首都师范大学;2014年
4 于佳平;基于核函数的极化SAR相干斑抑制研究[D];西安电子科技大学;2014年
5 张世吉;极化SAR目标检测算法研究及软件设计[D];西安电子科技大学;2014年
6 杜丽娜;极化SAR/InSAR信号处理与应用[D];西安电子科技大学;2014年
7 张晶晶;基于混合块相似性的极化SAR相干斑抑制研究[D];西安电子科技大学;2014年
8 李杨;基于Bootstrap统计方法的SAR图像相干斑抑制研究[D];西安电子科技大学;2014年
9 寇杏子;结合极化特征和图像特征的极化SAR图像分类研究[D];西安电子科技大学;2014年
10 刘佳颖;基于粒子群优化和Freeman分解的SAR图像分割与分类[D];西安电子科技大学;2014年
，本文编号：2233095
本文链接：https://www.wllwen.com/kejilunwen/wltx/2233095.html
