GF-2遥感影像城市空间信息提取与应用研究

发布时间：2019-03-15 18:29

【摘要】：遥感技术是对地观测,获取地面空间信息的主要技术手段之一。随着遥感技术的不断发展,高分辨率遥感影像的研究和应用越来越受到重视。高分辨率遥感影像的出现,“面向对象”影像分类方法随之崛起,围绕它的研究和应用由点及面地推广开来,运用于城市制图、城市土地动态变化监测、公共安全管理、突发情况预警等方面的研究,可以说面向对象图像分析(objectoriented image analysis,OBIA)已经在多个领域硕果累累。与传统的基于像元的遥感影像分类的方法相比,面向对象影像分类能克服“椒盐噪声”和“混合像元”等诸多问题,通过影像分割算法,将图像分成若干“图像对象”,借助于这些对象共有的几何特征、光谱特征、纹理特征和上下文关系等,能对每一类地物进行准确的提取,具有巨大的优势。但是目前利用面向对象分类的研究主要集中在森林、农业、城郊等地带,直接面向城市内部复杂地物提取的研究还比较少,且这些研究的研究区域都较小,未能充分利用这些“图像对象”的特征来提取复杂的空间信息,没有体现面向对象分类的优势。针对这些问题,本文从整体出发,以提取城市空间信息为目的,以北京市为研究区域,以1m空间分辨率的GF-2多光谱与全色融合影像为数据支撑,按照面向对象分类思想,以本文提出的V-I-W模型为理论指导,对主要的城市空间信息进行组织,将城市空间信息分为3大类14小类,通过多尺度分割、光谱差异分割来获取GF-2影像的“图像对象”,按照光谱特征、波段值特征、几何特征、纹理特征、和上下文关系建立各空间信息的提取规则,从而提取主要的城市空间信息,构建RS与GIS的桥梁,用于城市空间分析应用。本文的研究结果表明,城市空间信息提取的总体精度为83.24%,Kappa系数为0.8069。单一类别的地物提取方面,植被,河流提取精度较高,高于80%,城市道路,房屋等建筑的提取精度稍低,总体的空间信息提取结果符合预期。
[Abstract]:Remote sensing is one of the main technical means for Earth observation and acquisition of ground space information. With the development of remote sensing technology, more and more attention has been paid to the research and application of high-resolution remote sensing images. With the emergence of high-resolution remote sensing images, the object-oriented image classification method has emerged. The research and application of object-oriented image classification has been extended from point to area and applied to urban cartography, urban land dynamic change monitoring, public safety management, and so on. It can be said that object-oriented image analysis (objectoriented image analysis,OBIA) has already achieved great achievements in many fields, such as emergency early warning and so on. Compared with the traditional pixel-based remote sensing image classification method, object-oriented image classification can overcome many problems such as "pepper-salt noise" and "mixed pixel". Through image segmentation algorithm, the image can be divided into several "image objects". With the help of common geometric features, spectral features, texture features and context relations, each kind of objects can be extracted accurately, which has great advantages. But at present, the research of object-oriented classification is mainly focused on forest, agriculture, suburb and so on. There are still few studies directly facing the extraction of complex features within the city, and the research areas of these studies are small, and the research area of these studies is relatively small. The features of these "image objects" are not fully utilized to extract complex spatial information, which does not reflect the advantages of object-oriented classification. To solve these problems, this paper aims at extracting the urban spatial information, taking Beijing as the research area, taking the 1 m spatial resolution GF-2 multi-spectral and panchromatic fusion image as the data support, according to the object-oriented classification idea, and aiming at extracting the urban spatial information, taking Beijing as the research area. Under the guidance of the V-I-W model proposed in this paper, the main urban spatial information is organized, and the urban spatial information is divided into 3 categories and 14 sub-categories, and the urban spatial information is divided into three categories and 14 sub-categories through multi-scale segmentation. Spectral difference segmentation is used to obtain "image objects" of GF-2 images. According to spectral features, band value features, geometric features, texture features, and context relations, the extraction rules of spatial information are established. Thus, the main urban spatial information is extracted, and the bridge between RS and GIS is constructed, which can be used in urban spatial analysis. The results of this paper show that the overall accuracy of urban spatial information extraction is 83.24%, and the Kappa coefficient is 0.8069. The extraction precision of vegetation and river is higher than 80%. The extraction accuracy of urban roads and buildings is lower than that of other buildings. The overall spatial information extraction results are in line with expectations.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】