【作者】 徐丰；

【导师】 金亚秋；

【作者基本信息】 复旦大学 ， 电路与系统， 2007， 博士

【摘要】 在空间科学与空间应用的发展开始在全球各国复兴之时,在合成孔径雷达(SAR)开始在国民经济发展与国家安全保障密切相关的各遥感领域得到空前广泛的应用之时,作为本领域的博士学位研究论文,本文全面研究和发展了一套建立在电磁波与地表目标相互作用的物理基础之上、从散射传播机理与雷达成像原理出发、进行复杂地表结构的一维脉冲回波模拟和综合自然场景的二维雷达成像模拟、实现地表分类及参数反演和目标重建的基于极化合成孔径雷达的正向与逆向有机结合的遥感理论。在这套理论的研究过程中发展了一系列原创性的散射模型和散射估计方法、回波与成像模拟算法、极化信息解析理论、地表参数反演和三维目标重建理论与方法,旨在初步解决当前主动微波遥感特别是合成孔径雷达信息获取方面的部分问题,也为将来合成孔径雷达技术发展提供帮助和参考。全文分正向研究、逆向研究和扩展研究三部分。正向研究部分首先建立了粗糙面上非球形粒子层的矢量辐射传输模型,解得一阶Mueller矩阵解中包含面散射、体散射及相互作用的五项散射机制,用于模拟植被覆盖地表的极化散射,其中垂直地面的介质圆柱模拟树干,随机取向的细长圆柱模拟树枝,针状盘状粒子模拟针叶阔叶等。对比实验数据,该模型能有效估计植被地表的极化散射。通过不同算例的分析,研究了不同散射机制的贡献与植被层和地表面各物理参数的联系。在矢量辐射传输方程中引入时间变量并转换到频域求解,可进一步将该模型推广到时域,用于模拟时变入射波探测植被地表的极化回波波形。在此基础上发展了具有上下两层粗糙界面、中间镶嵌随机分布粒子的次表面介质层的模型,解得一阶Mueller矩阵解中包含七项散射机制。该模型延伸应用在脉冲波探测月球表面月壤层的问题。按月壤真实参数模拟低频脉冲波探测月壤层,发现回波波形能清晰反映月壤结构及物理特征等信息,据此提出了用低频段极化雷达脉冲波探测月壤层厚与层结构的新方案。在散射与回波模拟的基础上,继而研究了包含植被、建筑物及各种地形的复杂自然场景的SAR成像模拟。提出映射投影算法(MPA)以快速计算复杂场景的散射系数图,首先对场景进行网格划分,然后按入射方向依次累积各网格的衰减到投影平面、累加其散射到映射平面。该算法考虑了可穿透性植被目标的散射与衰减、树干及建筑物墙面与地面之间的多次散射、以及成像空间中任意位置地物的散射、衰减与遮蔽。由散射系数图经相干斑模拟、原始信号生成、成像压缩得到模拟SAR图像。对应地,逆向研究部分从零维散射信息或单个像素信息、一维回波信号到二维图像数据分别研究了极化信息解析和地表分类、回波信号处理和参数反演、多方位观测和三维目标自动重建三个主题。针对极化SAR的地表分类应用,提出目标全极化散射的去取向理论(De-orientation),通过对目标散射矢量的变换将目标旋转到一个特定的取向,使得原本取向不同但特征相同的目标能具有相同的散射信息,凸现目标本身特性。同时解析定义了去取向参数:去取向角、投影介电极性、介电跳跃度等,再加上已有的熵,四个参数构建无监督地表分类模式,由极化SAR数据将地表分为如地面、海面、森林、农田、城区、郊区等各种类别,对于同一类地表又可以由去取向角分析地物的取向分布特征,达到进一步细化分类的目的。用星载与机载极化SAR数据进行分类实验,对比传统的Alpha-A-H分类方法,结果表明去取向分类方法具有更好的分类性能。在对植被地表回波波形的分析中发现:波形传递了植被层与地表面的各种信息。由此,建立散射系统响应函数模型,将散射过程看成一个由入射波信号激励的系统,而回波就是系统响应。参考时域Mueller矩阵解的表达式,设计了植被地表这一散射系统的响应函数解析式,该响应函数由四组系统参数描述。用自适应非线性最小二乘法去拟合回波波形,可以成功估计这些系统参数。进一步通过推导系统参数与物理参数的理论关系,再结合先验知识,发展了同时反演粒子层厚度、粒子尺寸、取向、密度、介电常数、地面粗糙度和湿度的方法。据此提出了用低轨飞行的植被穿透脉冲雷达来获取植被与下垫地面信息的新方案。最后提出了用多方位SAR图像重建三维建筑物目标的方法。分为目标像的提取和多方位重建两个部分。目标像的提取流程包括先用恒虚警率检测器检测各目标像的边缘,并用脊滤波细化,再用平行线Hough变换检测平行线段,最终得到平行四边形建筑物像。多方位重建时,首先用统计几何描述不同方位的建筑物像,建立多方位目标像的参数相关模型给出其极大似然估计,再设计自动重建算法,从多方位检测结果中以较高的可靠性重建建筑物目标。用模拟和真实的SAR图像进行的重建试验,结果证明该方法的有效性,并给出发展多方位星载SAR重建应用的初步建议。扩展研究部分包括双站SAR研究和目标与环境散射估算。在双站SAR的研究中,首先针对平飞斜视配置,推导了双站SAR信号模型的频域表示式,给出了分离距离关联的近似解和较为精确的迭代解,对应地给出近似解和迭代解两种成像压缩算法。同时还将距离多普勒算法推广到双站情况。最后通过成像模拟说明了它们的有效性和优缺点。然后,在单站SAR映射投影算法基础上,通过三维投影和映射实现双站SAR成像模拟的映射投影算法(Bi-MPA),用于模拟复杂自然场景的双站SAR图像。通过设计各类场景并模拟双站SAR图像,分析了双站SAR图像上不同地物的成像特征与极化特征,发现传统极化特征参数在双站SAR图像上不再能有效表现地物散射的极化特征。对此提出统一双站极化基变换,并修改了极化特征参数的定义,使其保持原有的分离取向关联等优点。结果表明,统一双站极化基变换后,不同地物散射的极化特征更明显,而重新定义的极化特征参数也能反应不同散射机制,提供了BISAR图像解译和地表分类的初步手段。针对目标与环境的复合散射问题提出双向解析射线追踪法(BAR-Tracing),对于由面元和边缘建模的复杂目标,在入射方向和散射方向的反方向分别追踪射线并记录散射元上照射到的各阶射线,由此计算所有由一次散射和多次反射构成的多次散射项。其中,引入面元反射和散射发生概率以避免重复计算面元的散射贡献。此外,还用包含相干反射分量和非相干漫散射分量的粗糙面元建模粗糙面环境,实现目标与环境共存问题的通用解法。该技术使得大面元无需细剖分,从而降低复杂度。通过二面角、海面船舰等算例与精确的计算电磁学方法的对比,表明该方法能够提供较为可靠的散射快速估算。由正向研究、逆向研究和扩展研究构成的这套全极化合成孔径雷达遥感理论是作者四年多博士研究工作的总结,期望能初步解决合成孔径雷达遥感的部分问题。其间还有许多有待完善和深入的地方,何况技术和需求的日新月异,这都要求继续开展更深入更广泛的研究。更多还原

【Abstract】 SYNTHETIC APERTURE RADAR (SAR) has become one of the most important tools of earth observation and monitoring, attributable to its merits of day-and-night and all-whether operation and the capability of high-resolution imaging. In recent decades, SAR related techniques, e.g. interferometry, polarimetry, bistatic and multistatic SAR, polarimetric interferometry, and applications, e.g. topographic information retrieval, terrain classification, urban planning, military surveillance, forest management, have been extensively studied and developed.Research of SAR remote sensing can be categorized as electromagnetic scattering modeling, imaging and raw data compression, scattering signature interpretation and parameter inversion, image processing, etc. The purpose of SAR imaging is to acquire the information we need for applications in different areas. It has to resort to corresponding inverse theories and then extract the information delivered in SAR imagery. However, most inverse theories are based on direct analysis including theoretical modeling and simulation. It goes without saying that both direct studies and inverse approaches are very important and mutually beneficial.During the past several years of my doctoral research, I started from basic theories of electromagnetic scattering modeling for different types of terrain surfaces, rough surface to vegetation canopy, urban buildings to complex targets, and developed novel theories and approaches to both direct modeling and simulation and inverse retrieval and interpretation, for the applications on various circumstances, like vegetated areas, urban environments, ocean surfaces etc. The dissertation is organized as three main parts, the direct research, the inverse research and the extended research.The first topic of direct research is scattering modeling for vegetated terrain surfaces. A radiative transfer model for a layer of randomly oriented non-spherical particles with underlying rough surface was developed. Randomly oriented cylinders, disk-like and needle-like small particles are used for simulation of trunks, branches, deciduous and evergreen leaves, respectively, with underlying randomly rough surface for simulation of soil ground. It takes account of five scattering mechanisms of volume, surface scattering and their interactions. Numerical results match well with experimental data which shows the effectiveness of this model. It is useful to analyze and simulate scattering from vegetated areas and has been adopted in most of my research including scattering prediction, analysis and validation.The second topic of direct research is simulation of radar echoes from terrain surfaces. The abovementioned model is further extended to the time domain, by introducing the time variable into the radiative transfer equation and solving it in the frequency domain. Based on this, a more complicated model for a subsurface media layer with randomly distributed particles embedded in the middle and both rough upper and bottom interfaces is developed. It is used to simulate polarized echoes from lunar regolith for the lower frequency pulse radar detection. It is observed from the simulation that information of regolith structure and properties are hiding in the varying pattern of echo profiles. Therefore, a new mode for lunar exploration is proposed, using low orbit wideband pulse radar.A more advanced level of direct study should be the imaging simulation based on fundamental scattering models. Regarding the fact that natural scenes are more complicated including randomly distributed, penetrable or impenetrable objects, such as vegetation canopies, manmade structures and perturbed surface topography, a more adaptive simulation tool was developed. It takes account of scattering, attenuation, shadowing, foreshortening, overlay and multiple scattering of spatially distributed volumetric and surface scatterers, where a novel Mapping and Projection Algorithm (MPA) is devised to speed up the simulation of the entire process of scattering, extinction, mapping and projection in association with grid partitioning of the three-dimensional terrain scene. It also includes speckle simulation and raw data generation. It can simulate medium-resolution SAR images of comprehensive terrain scenes, which could have rivers, hills, urban and suburban buildings, timberlands, and farmland crops. It will be very useful in evaluating, interpreting and validating real SAR imagery.The inverse studies were conducted sequentially as the dimension of information increasing from oD scattering information to 1D echo signal, and to 2D SAR image. At the level of oD polarimetric scattering information within one single pixel, radar polarimetry and target decomposition theorems play important roles in the interpretation of polarimetric SAR images and its application on terrain surface classification. A novel de-orientation theory is proposed for the analysis of polarimetric scattering targets, plus a set of de-orientation parameters is defined with their respective physical meanings, namely the de-orientation angle, the projected dielectric polarity and the dielectric skippness. Together with the well-know entropy, a new unsupervised terrain surface classification scheme is proposed. Through theoretical simulation and real data experiments, it is demonstrated that the new parameters possess a higher degree of separation of terrain attributes than conventional methods.In order to utilize the lD information of the varying pattern of echo profiles, a novel multi-parametric inversion approach both for vegetation canopy and underlying ground is developed. As echo profiles provide rich temporal information showing the process of wave propagation and scattering through the complex media, those might identify contributions from different scattering layers or different volumetric and surface scattering. By treating the scattering process as some sort of signal excited system, the scattering can be described by a system response function which is explicitly expressed using a set of system parameters. These system parameters are estimated by fitting the echo profiles and they are further used to inverse multi-parameters of both vegetation canopy and the underlying ground based on their theoretical relationships.In the stage of high-resolution SAR applications, what we are concerned about is not an indicator parameter of the observed area, but rather the detailed information of the target, either in geometric or physical aspects. HR SAR images actually provide us with the possibility for extracting the specific parameters of an imaged target from the HR scattering object-image, for example, to invert the 3D sizes and position of a simple target, i.e. 3D reconstruction. Multi-aspect observations are especially important for 3D reconstruction due to the ambiguity of 3D objects in SAR images at one single aspect.An automatic method for detection and reconstruction of 3D objects from multi-aspect SAR images is proposed and realized with four steps. First, as a prior knowledge, the imaging features of the object are first generalized, such as geometric profile and spatial distribution. Then, to identify and extract the scattering image of the object from the SAR image, and the delivered information is inverted. Thereafter, the statistical description of the extracted object-image and its coherency to those of other aspects are provided. Finally, an automatic algorithm is designed to match object-images of different aspects in order to reconstruct the object. The potential application of this work is to effectively obtain the detail information of built-up areas from multi-aspect space-borne or airborne HR SAR data. It will be especially useful for urban planning and urban related research studies.Since the SAR techniques are developing rapidly, my research is also transfer to newly rising topics after I finished the direct and inverse studies. The first part of extended research is bistatic SAR. At first, bistatic image formation is studied. Under the translational invariant configuration of stripmap imaging BISAR, the frequency domain expression of signal model is derived. Two forms, respectively with an approximation of extracting the range dependence and an accurate iterative solution, are presented as well as their respective focusing methods. Meanwhile, the range-Doppler method is extended to the bistatic case. Imaging simulation of point targets shows the feasibility of these methods as well as their merits and drawbacks.Thereafter, by employing three-dimensional mapping and projection algorithm, imaging simulation of BISAR observation over complex scenario is realized. Some cases of simulated BISAR image are studied. Polarimetric characteristics of BISAR image are then discussed. It is found that some typical polarimetric parameters might become unable to describe scattering mechanism under bistatic observation. To circumvent this problem, a transform of unified bistatic polar bases for BISAR image is proposed. The conventional polarimetric parameters are redefined to retain the property of orientation independent in bistatic circumstance. Analysis of simulated images shows that the redefined parameters after the unified bistatic polar bases transform well describe different scattering mechanisms in BISAR imaging. It provides a primary tool for BISAR image interpretation and terrain classification.Scattering prediction of complex targets embedded in natural environments is the last topic of the extended research. Here, a new bidirectional and analytical ray tracing technique is proposed. Among the targets modeled by patches and edges, ray tracing is carried out both along the incident and inverse-scattered directions. By recording illumination and shadows information of rays at different orders, scattering of different orders can be calculated by using high frequency methods, such as physical optics and physical theory of diffraction. Besides, the concept of rough patch with coherent and diffused scattering parts is introduced to model rough surface, so as to deal with the co-presenting problems of targets and environments. This technique possesses the advantage of an electrical-size independent complexity of computation. Its precision for scattering predication is proved by comparing to conventional methods of computational electromagnetics.The entire set of remote sensing theories of polarimetric synthetic aperture radar consisted by the direct studies, the inverse studies and the extended studies is a summary of my doctoral research. It is expected to provide primary solutions to part of remote sensing applications of synthetic aperture radar. However, many aspects and issues are still remained for further study, not mentioning the rapidly developing techniques of synthetic aperture radar.更多还原