基于单幅高分辨率星载SAR影像的交通灾害信息提取方法研究

发布时间：2017-12-27 09:50

本文关键词：基于单幅高分辨率星载SAR影像的交通灾害信息提取方法研究　出处：《中国科学院大学(中国科学院遥感与数字地球研究所)》2017年博士论文　论文类型：学位论文

【摘要】：遥感技术在突发性地质灾害的灾后应急救援中发挥着越来越重要的作用,然而,由于灾害发生区域在灾后常伴随有云雾、阴雨等天气,极大的影响了光学遥感影像发挥的作用。SAR影像具有穿透云层的特性,能够不受天气影响,对受灾区域进行观测。随着SAR影像分辨率不断提高,使得对于关键的交通灾害对象,类似于桥梁、机场、滑坡、堰塞湖、倒塌建筑区域等重要的交通相关受灾地物的监测成为可能,监测结果可为交通应急指挥部门等单位的辅助决策。近年,基于SAR影像对灾后信息提取成为研究的热点,其中大多基于震前震后影像,然而从实际应用角度出发,震前震后影像获取困难,提高针对单幅高分辨率SAR影像的交通灾害信息的提取水平,对于提升整个交通行业的灾后应急处置能力,有着重要的影响、价值和意义。本文针对实际应用中遇到的若干问题,将灾害监测目标根据几何特征(点、线、面)分成几类,利用TerraSAR和高分3号卫星数据,开展以下灾害目标信息提取工作:(1)针对SAR图像相干斑噪声严重的问题,研究一种自适应的SAR图像相干斑噪声抑制方法,基于支持向量回归(SVR)的非线性拟合,在图像的对数域中分析支持向量回归误差与噪声类型参数之间的关联性,使噪声抑制模型能够根据每幅图像自身不同的噪声分布进行滤波。(2)针对目前SAR图像分类算法中经常出现的“分类结果整体最优与局部最优”无法同时满足的难点,研究支持向量机(SVM)在灰度共生矩阵和Gabor滤波两种纹理特征空间下的SAR影像分类,再引入RGAC模型,获得能够兼顾全局最优与局部精度的SAR图像分类方法。(3)针对交通灾害信息提取的应用需求及目前灾后应急工作中存在的困难,以2阶段RGAC模型影像分类算法为基础,结合不同地物几何、结构特点,开展枢纽节点类(包括桥梁、机场)、面状区域类(包括滑坡体、堰塞湖、倒塌建筑区)两大类目标的信息提取和分析工作,并对提取结果进行验证与精度评估。本文创新点总结如下:(1)在降噪阶段,提出一种能够根据SAR图像中相干斑噪声分布变化进行自适应调整的滤波算法。(2)图像分类阶段,提出一种改进的2阶段RGAC模型。通过引入机器视觉领域的RGAC模型,针对SAR影像改善其模型的初值与分类器,提高SAR图像的分类精度,尤其是分类边界上的提取精度。(3)特征提取阶段,结合SAR图像的散射特征、待提取目标的几何特征以及改进后的RGAC分类算法,在单幅高分辨率星载SAR图像中展开了多个具有重要实际应用价值的目标提取,取得了较好的提取结果。(4)数据源方面,除了使用TerraSAR卫星数据,还用到了高分3号卫星数据。
[Abstract]:Remote sensing technology is playing an increasingly important role in post disaster emergency rescue for unexpected geological disasters. However, due to the frequent occurrence of cloud and fog and other weather after disaster, the role of optical remote sensing image is greatly affected. The SAR image has the characteristics of penetrating the clouds and can be observed in the affected area without the influence of the weather. With the resolution of SAR image is increasing, the traffic disaster monitoring object key, similar to the bridge, airport, landslide, dammed lake, collapsed buildings and other important areas of traffic related disaster features as possible, monitoring results for traffic emergency department decision-making unit. In recent years, SAR images of the disaster information extraction has become a hot research topic based on which are based on pre and post earthquake images, however, from the practical point of view, before the earthquake after the earthquake image acquisition difficulties, improve the extraction level of traffic disaster information to single high resolution SAR image, to enhance the ability of emergency disposal of traffic industry in post disaster it has an important value and significance, influence. Aiming at several problems encountered in the practical application of the disaster monitoring target according to the geometric characteristics (point, line, surface) are divided into several categories, using TerraSAR and the high 3 satellite data, to carry out the following work: target information extraction of disaster (1) aiming at the problem of serious speckle noise of SAR image coherent speckle reduction method SAR image of an adaptive, based on support vector regression (SVR) nonlinear fitting analysis, support vector regression correlation between error and noise type parameters in the logarithmic domain of the image, the noise suppression model according to the noise of each image itself different distribution filter. (2) according to the currently used SAR image classification algorithm in the classification results of whole optimization and local optimization can not meet the difficulties of support vector machine (SVM) in SAR image classification and gray level co-occurrence matrix and Gabor filter two kinds of texture feature spaces, and then introduces the RGAC model, to obtain both SAR image classification the method of global convergence and local accuracy. (3) according to the application demand and the current traffic information extraction disaster disaster emergency work of the difficulties, the classification algorithm of 2 stage RGAC model based image, combining geometric and structural characteristics of different objects, to carry out the hub node class (including bridges, airport), surface area (including landslide, collapse and dammed lake construction area) information extraction and analysis work of the two categories of objectives, and verify and assess the accuracy of extraction result. The innovation points of this paper are summarized as follows: (1) in the phase of noise reduction, a filter algorithm which can adjust adaptively according to the variation of the speckle noise distribution in the SAR image is proposed. (2) an improved 2 phase RGAC model is proposed in the stage of image classification. By introducing the RGAC model in machine vision field, we can improve the initial values and classifiers of SAR models and improve the classification accuracy of SAR images, especially the accuracy of classification boundaries. (3) the feature extraction stage, combined with the scattering characteristics of SAR images, the extraction of RGAC classification algorithm and the improved geometric characteristics of the target, in a single high resolution spaceborne SAR images in multiple has important practical application value of the target extraction, achieved better extraction results. (4) in addition to the use of TerraSAR satellite data, the data sources also use high score 3 satellite data.
【学位授予单位】：中国科学院大学(中国科学院遥感与数字地球研究所)
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP75;X951;P694

【相似文献】