【作者】 陈强；

【导师】 匡纲要；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2010， 博士

【摘要】 作为电磁波四个基本特征之一,极化信息的开发利用对于提高雷达探测性能和完善雷达功能均具有重要作用。近年来,随着雷达极化测量技术和SAR成像技术的完善,各种星载和机载POLSAR系统不断涌现,极化信息的应用领域也不断拓展,但现有雷达极化理论尚不成熟、完善,已成为制约极化信息进一步开发利用的瓶颈。为此,本文以完善雷达极化理论和推进实用化为目的,对雷达极化中的目标特征极化、散射建模等若干理论问题进行了深入、系统地研究,并在大量高分辨POLSAR实测数据的支持下,采用理论分析和实验验证的研究方法,讨论了这些理论在POLSAR图像地物分类中的联合应用。论文工作的开展基本遵循一致的步骤,即在充分了解前人研究工作、现关成果及存在问题的基础上,找准论文的切入点展开深入的研究。归纳起来,论文主要开展的工作有:1)发展和完善了现有相干情形目标特征极化理论。推导了天线接收功率全局最大值及其极值条件,针对现有研究未涉及目标参数变化对特征极化的影响,首次结合功率密度图研究了这一问题,并得出了一些重要的结论,为简化目标特征极化求解、预判目标特征极化位置提供了理论支撑。该方面的研究成果已被《中国科学E辑:技术科学》中英文刊用;2)以实用化为目的,对现有单变量Lagrange乘因子求解算法进行了优化,提高了非相干、同极化通道特征极化求解效率。单变量Lagrange乘因子求解实质归结为一个以Lagrange乘因子为自变量的一元六次方程求解,其求解面临两个问题:a)天线接收功率极值与该方程根之间对应关系未知,使得求解过程搜索不必要的根,增加了算法运算量;b)方程根隔根区间未知,造成需人为确定每个根的隔根区间,不利于算法工程实现。为此,文中从两方面入手开展研究:其一,理论分析天线接收功率极值与该方程根之间对应关系,并得出了“天线接收功率最大、最小值分别对应该方程最大、最小根”的结论;其二,理论推导方程最大、最小根的隔根区间。以此为基础,最后给出了优化后的单变量Lagrange乘因子求解具体步骤。该方面的研究成果已连续在《信号处理》、《IEEE Transactions on Geoscience and Remote Sensing Letter》等刊物上发表;3)在非相干、任意通道约束特征极化方面,统一了任意通道约束天线接收功率数学模型,从而将不同通道约束特征极化求解归结为同极化通道天线接收功率形式统一求解。利用变极化基技术对该模型进行了简化处理,以此为基础,提出了一种基于区间定位的目标特征极化求解新算法。该算法在理论确定天线接收功率极值在功率密度图上的单调区间基础上采用区间二分法迭代搜索极值位置。实验结果和理论分析表明,该算法较优化单变量Lagrange乘因子求解更具有运算优势。该部分研究成果已在《测绘学报》上发表;4)以实用化为目的,提出了一种基于“三步”解耦思想的特征极化求解新方法,解决了非相干、无约束特征极化求解困难的问题。将“三步”解耦思想应用于分布式目标面临两个新问题:a)分布式目标的散射回波始终是部分极化的,而部分极化波又只能采用Stokes矢量表征;b)部分极化波的天线失匹配接收不能简单的将天线极化与入射波正交或相同,因为天线极化为完全极化的。文中首先分析了部分极化波的最优接收问题,得到了任意部分极化波天线接收功率上下限和失匹配接收条件,并指出天线接收功率上下限实质是发射天线极化的函数。以此为依据,给出了非相干、无约束特征极化求解步骤:a)通过天线接收功率上下限函数优化得到发射天线最优极化,即对应天线接收功率下限最小,或上限最大的发射天线极化状态;b)利用散射方程计算最优散射回波;c)利用部分极化波失匹配接收条件得到接收天线最优极化;该部分研究成果已在《信号处理》上刊用;5)在详细分析现有目标散射建模研究成果的基础上,得出了“基于H/Alpha的散射描述是现有目标散射描述中最合理的一种”这一结论。但通过对极化散射熵的理论分析发现,H/Alpha散射描述并不总是合理,因为它假设目标总存在占绝对优势的平均散射机制并不总是成立。合理的描述应是对于不同散射随机性的目标,采用不同方式表征其散射:对于低散射随机性目标,可采用主散射机制表征;对于中散射随机性目标,应采用主散射机制和次要散射机制联合表征;对于高散射随机性目标,其散射已无法采用某种散射机制表征。这一研究为后续的散射特征提取和散射分类提供了理论依据。该方面研究成果已被国际权威杂志《IEEE Transaction on Geoscience and Remote Sensing》建议小修后刊用,获得了审稿专家很好的评价;6)发展和完善了散射相似性理论。针对现有散射相似性无法应用于分布式目标的局限,定义了一种度量目标散射相似性的新参数。作为一种实际应用,利用该参数提取了目标与金属球、二面角等典型目标的散射相似性参数,并指出:球面散射相似性、二面角散射相似性和体散射相似性具有双重物理内涵,既表征了目标与典型目标散射相似程度,又表征了这些典型散射对目标后向散射贡献率。着重分析了球面散射相似性独特性质,及与极化散射熵、平均Alpha角之间的数学约束关系。这些研究为后续散射分类应用提供了理论支持。该部分研究成果已连续被《IEEE Transaction on Geoscience and Remote Sensing Letters》、《信号处理》、《电子学报》等刊用;7)在分析现有散射随机性度量参数的基础上,得出“极化散射熵是目前度量目标散射随机性最合理的参数”这一结论。针对极化散射熵存在不足,着重开展了两个层次的研究:其一,鉴于极化散射熵采用相干矩阵特征值定义,在分析相干矩阵不变特征量基上,定义了基于2-范数的极化散射熵替代参数,并分析了它与极化散射熵之间的数学约束关系;其二,考虑到散射回波极化度也包含了散射随机性信息,定义了一种基于极化度的散射随机性度量新参数。结合典型粒子云团,从理论上阐明了新参数度量目标散射随机性的合理性。实测数据实验结果表明,上述参数克服了极化散射熵的运算量偏大问题。该方面的研究成果部分已被《信号处理》刊用;8)对POLSAR图像相干斑抑制技术进行了全面地研究,指出了现有类多视平均处理一类相干斑抑制算法在运算量、目标信息保持等方面的不足,并提出了一种基于像素筛选的POLSAR图像相干斑抑制新算法。该算法避免了现有相干斑抑制技术无法同时兼顾相干斑抑制和目标极化、边缘等信息保持的局限,能够自动地筛选同主散射机制、均匀区像素参与最小线性均放误差滤波处理,为POLSAR图像相干斑抑制提供了新的思路。该方面研究成果已被《信号处理》刊用。9)针对现有散射分类存在的不足,从两个层次开展了研究:其一,提出了一种基于球面散射相似性和极化散射熵替代参数的H/Alpha替代分类方法,克服了H/Alpha散射分类运算量偏大的问题,但仍存在类别边界人工选取的问题;其二,提出了一种基于多散射相似性参数和散射随机性的散射分类新方法。该方法采用散射随机性将目标分为高、中、低散射随机性三大类,然后利用球面散射相似性、二面角散射相似性和体散射相似性提取目标散射机制鉴别表征目标散射,即对于低散射随机性目标,采用散射相似性最大对应的典型散射表征目标散射;对于中散射随机性目标,采用散射相似性较大对应的典型散射联合表征目标散射;对于低散射随机性目标,不进一步细分。为POLSAR图像散射分类提供了新的思路和途径;该方面研究的部分成果已被《信号处理》、《电子学报》等刊用,部分成果已投国际权威杂志《IEEE Transaction on Geoscience and Remote Sensing》;10)针对Wishart距离度量运算量偏大的不足,提出了一种度量两类目标散射差异性的新参数,分析了目标SCR变化对该参数影响,进而阐明利用目标特征极化理论提高目标SCR的重要性。以此为基础,又提出了一种POLSAR图像地物无监督迭代分类新方法。即将散射分类结果作为初始类别,利用新参数进行类别迭代调整,取得了较好的实验效果。该方面部分研究成果已被《电子学报》等刊用,部分成果已投国际权威杂志《IEEE Transaction on Geoscience and Remote Sensing》;更多还原

【Abstract】 As one basic feature of electromagnetic wave, the utilization of radar polarization has the advantages of improving radar detection and recognition ability. Thanks to advances in POLSAR systems and information technology over the past decade,radar polarimetry has become a subject of recurring and globally intensifying interest all around world. However, the basic theories of radar polarimetry are still not perfect, which blocks its applicactions. With such a researching background, this thesis comprehensively investigates multi basic theories of radar polarimetry, namely target characteristic polarization theory and target scattering modeling. Some new theories and techniques are proposed, and their joint applications in improving terrain classification of polarimetric SAR(POLSAR) image are also systematically studied by theoretic analysis and experimental validation with lots of real high-resolution POLSAR data.Our research follows the ordinary process: first, existing work is thoroughly reviewed; second, based on related research and problems, researching points of this thesis are determined and further research is carried out. The work of this thesis can be concluded as follows.1) Target characteristic polarization theory for the coherent case is studied in greater detail, which develop target characteristic polarization theory ulteriorly. Power extrema and their conditions are derived firstly. And then the behavior of target characteristic polarization states as functions of target parameters are discussed for the first time and important conlusions are gived, which can be used to decrease computational complexity of target characteristic polarization states and anticipat their positions on the power density plot. Some of this studies have been published in some Chinese journals such as Signal Processing, cience in China Series E: Technological Science, etc.2) An improven Lagrange multiplier method is proposed to find the optimal polarizations for the co-polarized incoherent case. The traditional Lagrange multiplier method requires solving a sixth-order polynomial equation of the Lagrange multiplier. However, since it is unknown which of the roots corresponds to the maximum received power and which corresponds to the minimum, all the real roots need to be found, which will cause heavy computational burden. Thus two aspects are studied in thesis. First, We have proved for the first time that the largest and smallest roots of the Lagrange multiplier equation correspond to the maximum and minimum received powers, respectively. Second, the minimum search intervals of these two roots are also given by some theoretical analysises. Based on these researches, the flow of the proposed procedure is presented, which takes at least one to three times less computational time than the original procedure. The achievement of this aspect has been accepted to publish by“Signal Processing”and IEEE Transactions on Geoscience and Remote Sensing Letter.3) A fast algorithm is proposed to find the optimal polarizations for arbitrary polarized channel. Firstly, the function of received power in arbitrary channel is unified as the form of copolar power. Then based on the change of polarimetric basis, the copolar power is analyzed theoretically to obtain the minimum interval of target optimal polarization state, which provides theoretic support for simpling the process of obtaining optimal polarization states or anticipating their positions. Finally, the interval dichotomy is used to search in the minimum interval of target optimal polarization state. The experiment results have demonstrated the validity of the proposed algorithm. The achievement of this aspect has been accepted to publish by“acta geodaetica et cartographica sinica”.4) A three step procedure is proposed to find the optimal polarizations of random targets for the case of no restriction between receiving and transimitting antenna. Considering that the scattering waves of random targets are partially polarized, the optimal receiving is firstly discussed for partially polarized wave, and the mismatched/matched receiving conditions and their corresponding receiving powers are obtained. Then the flow of the proposed three procedure is given: a) the optimal transmitted polarization states is obtained by optimizing mismatched/matched receiving power; b) the optimal scattering wave can be derived by substituting the optimal transmitted polarization states into the scattering equation; c) the optimal received polarization states can be obtained by the mismatched/matched receiving conditions for the incoherent case. The experiment results have demonstrated the practicability of the proposed procedure. The achievement of this aspect has been accepted to publish by“Signal Processing”.5) after thoroughly reviewing existing works, we conclude that the scattering model based H/Alpha is the most integrated one of target scattering now. But through some theoretical analysis of polarimetric scattering entropy, we find that this scattering model is not reasonable for some cases because it sets out with the assumption that there is always a dominant“average”scattering mechanism, which doesn’t always come into existence. And the more reasonable model is to represent target scattering with different scattering mechanism according to target scattering randomness. That is to say, the main scattering mechanism is used for low scattering randomness; the main and minor scattering mechanism are adopted together for median scattering randomness. The achievement of this aspect has been submitted to IEEE Transactions on Geoscience and Remote Sensing.6) Aiming at the deficiency of the similarity parameter to measure the degree of scattering similarity between distributed targets and canonical targets, a new parameter, namely polarimetric scattering similarity, is proposed based on the Pauli vector of a canonical target and the coherency matrix of a distributed target firstly. And the existing similarity parameter is just a special case of the proposed parameter. As an application, the parameters to several canonical targets, eg. metal sphere are derived and their physical meanings or properties are discussed. Finally, the relationship between the surface scattering similarity and average apha angle or polarimetry scattering entropy is also analysized. The achievement of this aspect has been accepted to publish by“Signal Processing”,“Acta Electronica Ainica”and IEEE Transactions on Geoscience and Remote Sensing Letter.7) Aiming at the deficiency of the present parameters to measure target scattering randomness, a novel parameter is defined based on the degree of polarization of target scattering wave. Firstly, after the analysis of the relationship between the degree of polarization of target scattering wave and target scattering randomness, the novel parameter is constructed with the degree of polarization of target scattering wave. And then the feasibility of the novel parameter to measure target scattering randomness is clarified by the analysis of the behavior of the novel parameter as a function of the scattering order in volume backscatter or target anisotropy. As the new parameter involves simple operation, it overcomes the time-consuming operation of the present parameters. The achievement of this aspect has been accepted to publish by“Signal Processing”.8) A fast algorithm based on the automatic censoring is proposed to filter the speckle in POLSAR image. Firstly, a rectangle widow is sliding through each pixel of POLSAR imagery. For each current pixel, a global threshold is used to determine whether the current pixel is a point target. If not, then the homogeneous pixels of the same main scattering mechanism with this current pixel are censored automatically in the current sliding-window. Finally, the locally linear minimum mean-squared error (LLMMSE) estimator is used to filter this pixel with the selected pixels. The selection of homogeneous pixels based on the ML texture reduces the effect of speckle; the selection of the same main scattering mechanism pixels based on scattering similarity preserves the main scattering mechanism. The experiment results show that the proposed algorithm has the well theoretical and actual performance in speckle filtering, texture preservation and target main scattering preservation. The achievement of this aspect has been accepted to publish by“Signal Processing”.9) Aiming at the deficiencies of the H/Alpha based classification, an alternative to H/Alpha based classification is proposed by replacing alpha angle with surface scattering similarity. But it still has the deficiency of arbitrarily fixed linear decision boundaries. Thus, a new scheme of unsupervised terrain classification is proposed based on multi scattering similarities. This new scheme divides terrain scattering into three classes with target scattering randomness firstly; and subdivides three classes into ten classes with surface scattering similarity, double scattering similarity and volume scattering similarity ulteriorly. As terrain scattering class is determined by target scattering similarities automatically, the deficiency of the arbitrarily fixed linear decision boundaries of Alpha angle is overcomed. Surface scattering, double scattering and volume scattering are the inherent characteristics of terrain physical scattering mechanisms, the classification result of the new method is consistent with the real terrain scattering. The achievement of this aspect has been accepted to publish by“Signal Processing”and“Acta Electronica Ainica”.10) Aiming at the deficiencies of wishart classifier, a new parameter is proposed to measure target scattering difference first, and then applied to refine the result of the above proposed scattering classification. And good performance is avcheived for POLSAR image terrain classification. The achievement of this aspect has been accepted to publish by“Acta Electronica Ainica”.更多还原