综合极化信息和改进FNEA的PolSAR影像建筑物提取

发布时间：2018-04-19 02:38

本文选题：面向对象 + 建筑物提取　；参考：《山东科技大学》2017年硕士论文

【摘要】：极化合成孔径雷达(Polarimetric Synthetic Aperture Radar, PolSAR)是在原有单极化合成孔径雷达基础上发展来的能接收不同极化散射通道散射信息的新型雷达。极化SAR获取到的不同极化通道的极化信息反映不同类型地物对不同极化方向雷达波的响应,研究建筑物在不同极化通道下的响应,可以提供与光学信息截然不同的另一视角来指导建筑物的提取。然而,目前利用极化信息开展的建筑物研究多是基于像素级的,面向对象的研究较少。本文重点研究面向对象的极化SAR建筑物提取中的关键步骤——分割和提取。首先针对分形网络演化(Fractal Net Evolution Approach,FNEA)方法应用于极化SAR影像分割时出现边缘漂移现象的问题,引入边缘特征对分形网络演化方法中的底层像素合并进行引导;接着针对保持极化散射特性的建筑物提取方法利用特征单一、建筑物提取精度较差的问题,引入极化方位角特征、亮度特征和纹理特征减少建筑物的漏检和误检;并通过机载全极化数据设立三组对比试验,证明了改进的分割方法和建筑物提取方法的有效性。论文的主要研究内容及创新点如下:(1)从描述电磁波极化特性的Jones矢量、极化椭圆和Stokes矢量出发,总结了目标的极化散射特性,并比较了几种极化SAR影像相干斑滤波方法的效果,为后续的研究打下基础。(2)提出改进FNEA的极化SAR影像分割方法。针对FNEA方法应用于极化SAR影像分割时因初始合并边缘信息缺乏导致分割结果中出现边缘漂移现象的问题,引入边缘特征引导像素级合并生成初始对象,以该初始对象为起点,计算融入极化信息的广义相似度准则,搜索最优对象合并,完成多尺度阈值下的分割,并以单特征建筑物提取的总体精度作为尺度参数的筛选标准,得到最优尺度的分割结果,用于后续对象级建筑物提取。(3)提出综合多特征的极化SAR影像建筑物提取方法。针对现有的面向对象的极化SAR建筑物提取方法利用特征较单一的问题,详细分析了漏检、误检形成的原因和对应的极化分解特征的表征,接着详细论述了综合利用极化方位角特征、亮度特.征和纹理特征进行改进的方法,并对方法实现过程中的若干关键点进行了讨论,同时,提出了对象占优因子对传统散射类型占优的方法进行了改进,形成了一套面向对象的综合多特征的极化SAR建筑物提取方法。
[Abstract]:Polarimetric synthetic aperture radar (Polarimetric Synthetic Aperture Radar, PolSAR) is in the original single polarization synthetic aperture radar based on the development of new radar can receive different polarization scattering channel scattering information to get to the SAR. The polarization in different polarization channel polarization information to reflect the different types of objects of different polarization radar wave response response research the buildings in different polarization channel, extraction can provide another perspective different from optical information to guide the building. However, current research by using polarization information of buildings is based on pixel level, research on object oriented segmentation and extraction less. Key steps of this paper focuses on the research of polarization SAR building object-oriented extraction at first according to fractal net evolution (Fractal Net Evolution Approach, FNEA) method is used in the segmentation of polarimetric SAR images When the edge drift problem, introduce the edge pixels with low-level features of fractal network evolution method in the guide; then in order to keep the polarimetric scattering characteristics of building extraction methods using single feature, the poor performance of building extraction, the polarization azimuth characteristics, brightness and texture of buildings to reduce missed and false detection; and through the airborne polarimetric data set up three groups of contrast experiments, proved the effectiveness of segmentation methods and building improved extraction method. The main research contents and innovations are as follows: (1) from the Jones vector to describe electromagnetic wave polarization, polarization ellipse and vector of Stokes, summarizes the polarimetric scattering characteristics of targets. And a comparison of several polarimetric SAR image speckle filtering method, lay the foundation for further study. (2) proposed the improved FNEA polarization SAR image segmentation Methods. The edge drift phenomena in FNEA method applied to polarimetric SAR image segmentation by merging initial edge information due to a lack of segmentation results in problems, using edge features to guide the pixel level combined to generate the initial object, the initial object as a starting point, calculate the generalized similarity criteria into the polarization information, search the optimal object with multi scale, complete threshold segmentation, selection criteria as the scale parameter and the overall accuracy in single building feature extraction, optimal scale segmentation results for subsequent object extraction of buildings. (3) extraction method is proposed for polarimetric SAR image building integrated multi features. Using the method of single feature extraction in existing object-oriented. Polarized SAR buildings, a detailed analysis of the detection, characterization of Decomposition Characteristic polarization error causes and the corresponding form, then in detail Comprehensive utilization of polarization azimuth characteristics, brightness. A method of improving signs and texture features, and some key points in the implementation process of the method is discussed, at the same time, put forward the method of object dominant factor on the traditional dominant scattering type was improved, forming a multi feature set of object oriented polarization SAR building extraction method.

【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】