作　　者： ; ; ; ;

机构地区： 国家海洋局第三海洋研究所

出　　处： 《地理与地理信息科学》 2013年第3期35-38,47,F0002,共6页

摘　　要： 利用高分辨率遥感数据,针对"512"地震中都江堰紫坪铺镇震害损毁建筑物的特征,研究了基于像元和面向对象两种快速提取方法。基于像元提出了基于"多层次区域分割"的震害损毁建筑物提取方法,与传统的分类提取方法相比,它体现着面向对象分析方法中的多层、分割思想。面向对象,利用多尺度分割思想,按照分割尺度的大小建立了三层体系结构,根据每一层的特征提取不同地物类别,分析其光谱、纹理、形状、上下文关系、空间位置等特征,建立各自的模糊判定规则,进行损毁建筑物的提取。分别从提取精度、目视效果和方法原理三方面对两种提取结果进行综合比较,结果显示:两种方法均能快速、准确地提取震害损毁建筑物信息,而面向对象提取方法成功地抑制了由于单个像元光谱异质性大而导致的高光谱遥感分类"胡椒盐"噪声,有效地解决了基于像元方法中存在的"同物异谱,同谱异物"现象,其总体精度(90.38%)比基于像元方法的总体精度(76.84%)更高。 According to the characteristic of earthquake collapsed building in Zipingpu Town of Wenchuan earthquake, this paper validated two kinds of fast extraction method of the collapsed building information based on pixel-oriented and object-oriented theories. The pixel-oriented method on the basis of the multi-layer regional segment idea which embodies the thoughts of layer and segment from object-oriented was applied. The key idea is to mask the other information besides collapsed building from im- age layer by layer. As compared to a traditional way, it is an innovative experiment because the computer processing becomes faster and faster in the end. When it comes to object-oriented method, the multi-scale segment algorithm was applied in order to build up three-layer hierarchy. Then by analyzing the spectrum, texture, shape, location and context of individual object-classes in different layer, the fuzzy determined rule system was established for earthquake collapsed building extraction, The two results were compared from three aspects which are precision assessment, visual effect and principle. The result showed that the two methods both can extract the earthquake collapsed building information quickly and accurately. The object-oriented result, suc- cessfully overcome the phenomenon of pepper salt noise caused by the spectral diversity of high-resolution remote sensing data and solved the problem of same object different spectrums as well as same spectrum different objects, achieved more scientific and accurate result with the overall accuracy of 90. 38% as compared to 76.84% of pixel-oriented method. It is concluded that the object-oriented image analyzing method will have an extensive application foreground in earthquake disaster information extraction of high-resolution remote sensing.

关 键 词： 基于像元 面向对象 间接提取 多尺度分割 模糊分类

领　　域： [自动化与计算机技术] [自动化与计算机技术]