特征值非负约束下的基于模型的极化SAR分解研究

发布时间：2018-01-19 03:04

本文关键词： 极化合成孔径雷达极化分解散射机制散射模型散射机制分类　出处：《武汉大学》2014年博士论文　论文类型：学位论文

【摘要】：作为一种主动遥感方式,极化合成孔径雷达(PolSAR)具有全天时全天候工作能力,其分辨率一般高于普通的真实孔径雷达。最近几年,它开始在军事,测绘,农业,林业,地质等领域得到广泛应用。作为一种从PolSAR中提取信息的重要方法,极化分解,尤其是基于模型的非相干极化分解,是最近几年PolSAR领域内最活跃的方向之一。它可以获得不同散射机制的功率和其它参数,进而用于PolSAR影像分类,干涉SAR,极化相干斑滤波,土壤粗糙度和湿度估计等。自从Freeman和Durden提出三分量分解法后到现在,研究人员已经提出了二十多种基于模型的非相干分解法。这些方法虽然得到了不少成功的应用,但是普遍存在诸如不满足特征值非负约束,出现负功率值,高估体散射功率,对极化信息利用不充分,对地面散射采用相干散射模型进行模拟,不能描述去极化,以及难以有效区分森林和分布方向不平行于SAR方位向的建筑物等问题。一般采用真实数据对分解法进行验证,缺乏和真值的比较,难以定量评估分解法对各分量功率及其它参数估计的准确度。针对上述问题,本文首先创建了一个基于极化分解的模拟框架,模拟不同分量的参数,计算它们不基于反射对称的散射模型,得到功率加权后的相干矩阵。通过对模拟得到的相干矩阵利用不同方法进行分解,可以将分解结果与模拟参数进行定量比较。作者还挑选了不同散射机制主导的模拟数据,以更好地模拟真实情况。本文提出了两种高度自适应的分解法。这两种方法都进行去方位处理,应用特征值非负约束到螺旋散射和体散射参数的计算,采用Neumann自适应散射模型和双极子来描述体散射,选择能让体散射解释最多交叉极化功率的参数作为最优的体散射参数。但是第一种分解法不基于反射对称计算体散射参数(简称为RAVD),导致在一般情况下,螺旋散射和体散射不能解释所有的交叉极化功率。为此,采用Neumann模型描述主导地面散射以解释剩余的交叉极化功率,采用相干模型描述次要地面散射。而第二种分解法计算体散射参数时基于反射对称假设(简称为RSVD),使得在大部分以表面或双次散射为主的区域,体散射和螺旋散射能解释全部的交叉极化功率,再由van Zyl分解即可获得表面散射和双次散射的参数。但是在部分森林地区,少部分交叉极化功率不能由体散射和螺旋散射解释。在这种情况下,对观测到的相干矩阵进行三分量分解,其中体散射和主导地面散射均由反射对称的Neumann模型描述。如果上述分解不能取得合理结果,则进行三分量反射不对称分解。利用模拟数据和UAVSAR数据所做的实验表明,在绝大多数情况下,这两种方法可以匹配除T13外其它观测到的相干矩阵中的元素。如果进行三分量反射不对称分解,则有可能匹配除T13虚部之外的其它相干矩阵元素。RSVD避免了负功率值的出现,而RAVD的结果中,负功率值的比例也低于0.070%。两种分解法明显降低了对体散射功率的高估,估计各分量功率的准确度高于几种最新的特征值非负分解法。在大多数情况下,RAVD估计不同分量的方位角随机度和复散射系数的准确度优于RSVD,而且在它的结果中,森林和延伸方向不平行于SAR方位向的建筑物可以较为容易地区分开。但是在以表面散射或双次散射为主的区域,RSVD估计各分量功率的效果优于RAVD. 本文还提出了一种基于功率的非监督散射机制分类法。散射机制类被定义为不同主导和次要散射机制的组合。通过分析不同散射机制的特征以及两种散射机制混合时的特征,作者给出了一种基于极化特征和特征域分割的散射机制分类法。由于该分类法基本不依赖于极化分解,所以避免了对体散射功率的高估和特征值分解。该分类法的效率大大高于Wishart-H/alpha法和模糊H/alpha法,而且能够提供次要散射机制的类别。该方法可以用于PolSAR影像的快速分类,其分类结果可以作为更复杂的分类器的初始分类。它还可能用于简化基于模型的非相干分解。在利用模拟数据的实验中,该方法给出的Kappa系数为0.864。该方法识别主导散射机制的效果显著优于H/alpha法,Wishart-H/alpha法和模糊H/alpha法。UAVSAR数据的实验表明,该方法能够有效识别森林和城区的主导和次要散射机制。
[Abstract]:As an active remote sensing method, polarimetric synthetic aperture radar (PolSAR) has the ability to work all day long during the entire time, the resolution is generally higher than the ordinary real aperture radar. In recent years, it began in the military, mapping, agriculture, forestry, geology and other fields has been widely used. As an important method to extract information from in PolSAR polarization decomposition, especially non coherent polarization decomposition based on the model, the direction of recent years is one of the most active in the field of PolSAR. It can obtain different power scattering mechanisms and other parameters, and then used for PolSAR image classification, SAR interference, coherent speckle filtering, soil roughness and humidity estimation.
Since Freeman and Durden proposed the three component decomposition method after up to now, researchers have proposed more than 20 kinds of non coherent decomposition method based on model. Although these methods have many successful applications, but generally does not meet the problems such as the eigenvalue of nonnegative constraints, the negative power value, overestimate the volume scattering power of polarization information do not use fully, to simulate the ground scattering by coherent scattering model, can describe depolarization, and it is difficult to effectively distinguish between the forest and the distribution direction is not parallel to the SAR direction to the building and other issues. Generally used to verify the decomposition method for the real data, and lack of true value, it is difficult to quantitatively evaluate decomposition method to estimate each component power and other parameters accurately.
Aiming at the above problems, this paper firstly creates a simulation framework based on polarization decomposition, the simulation parameters of different components, they are not based on the calculated scattering model of reflection symmetry, get the power weighted coherent matrix. Through the decomposition of the correlation matrix by using different method of simulation, we can decompose and simulation results quantitatively comparison. The author also selected the simulated data leading to different scattering mechanisms, in order to better simulate the real situation.
This paper presents two kinds of highly adaptive decomposition method. The two methods are carried out to the azimuth processing, using eigenvalue calculation non negative constraints to spiral scattering and volume scattering parameters, using Neumann scattering model and adaptive dipoles to describe the scattering, scattering can make selection interpretation parameters of cross polarization power as the most the optimal parameters of the scattering body. But the first decomposition method is not based on computational reflection symmetry scattering parameter (RAVD), resulting in general, spiral scattering and volume scattering cannot explain cross polarization power of all. Therefore, the Neumann model was adopted to describe the dominant ground scattering to explain cross polarization power surplus, by using the coherent model describe the secondary land scattering. And second kinds of decomposition method based on the assumption of computational reflection symmetry scattering parameters (referred to as RSVD), which in most of the surface or double scattering. The area, volume scattering and spiral scattering can explain cross polarization power all the parameters can be obtained by van Zyl decomposition of surface scattering and double scattering. But in the part of the forest area, a small part of the cross polarization power cannot be explained by scattering and spiral scattering. In this case, the coherent matrix are observed three component decomposition, the scattering and scattering are dominant ground described by Neumann model. If the reflection symmetry decomposition cannot obtain reasonable results, then decompose the three component reflection asymmetry.
Using the simulated data and the UAVSAR data experiments show that, in most cases, these two methods, except T13 observed coherent matrix elements. If the three components of the reflection asymmetric decomposition, it may, in addition to T13, the imaginary part of other coherent matrix elements to avoid the emergence of negative.RSVD the power value, and the result of RAVD, the negative power value was lower than two 0.070%. decomposition method significantly reduces the overestimation of the volume scattering power, the power component estimation accuracy is higher than some of the new non negative eigenvalue decomposition method. In most cases, the RAVD estimates for different azimuth angle random component and the complex scattering coefficient is more accurate than RSVD, but also in its results, the forest and the extension direction is not parallel to the SAR direction to the building can be separated more easily. But in the area of surface scattering or double In the region where the scattering is dominant, RSVD estimates the power of each component is better than that of RAVD.
This paper also presents an unsupervised classification method based on scattering mechanism of power. A combination of scattering mechanisms are defined for different class leading and secondary scattering mechanism. Through the analysis of characteristics of different scattering mechanisms and characteristics of two kinds of scattering mechanisms are mixed, the author gives a classification of scattering mechanism features and domain segmentation based on the method. Because the basic classification does not depend on the polarization decomposition, so to avoid overestimation of the volume scattering power and eigenvalue decomposition. The classification efficiency is much higher than the Wishart-H/alpha method and fuzzy H/alpha method, and can provide a secondary scattering mechanism category. The method can be used for fast classification of PolSAR image classification, initial classification the results can be used as more complex classifiers. It may also be used to simplify the non coherent decomposition based on model.
In the experiments, the Kappa coefficient method is presented for the 0.864. method to identify the dominant scattering mechanism of the effect is much better than the H/alpha method, Wishart-H/alpha method and experiment show that the fuzzy H/alpha method of.UAVSAR data, this method can effectively identify the dominant forest and urban areas and the secondary scattering mechanism.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P237;P225.1

【参考文献】