【作者】 陈小良；

【导师】 程烨； 付梅臣；

【作者基本信息】 中国地质大学（北京） ， 土地资源管理， 2009， 硕士

【摘要】 随着IKOONOS、QuickBird的普及以及WorldView-1的出现,高空间分辨率影像成为研究土地利用、覆被变化的最主要数据源。高空间分辨率具有海量的空间信息,清晰地纹理信息,传统的分类方法不能有效地提取这些里面的空间信息。基于象元的传统分类方法在分类中会出现胡椒盐现象,出现无效图斑,不能保证分类精度。人们将影像分割技术应用到土地利用覆被分类、地物信息提取之中发展成为今天的面向对象分类方法。面向对象分类以对像为基本处理单位,通过尺度、形状/颜色、紧致度/光滑度等参数界定对象,并在分割的基础上进行分类。面向对象的专业软件Definiens提供了很多种分割算法,包括四叉树分割、棋盘分割等,其中最普及的是多尺度分割。但是,在多尺度分割里的尺度参数选择一直困扰着学者们,有学者进行了这方面的探索,建立了最大面积法等准则,但是其太复杂,因此更多的研究者更乐意通过目视选择自己认为的最优尺度。本文作者通过建立同质性指数和异质性指数,对实验区选择了四个层次的尺度,并建立了包括道路、河流、有林地、旱地、水浇地、房屋、灾毁地、阴影、丛树、荒草地、临时篷房在内的土地利用覆被系统,在分割的基础上,选择最邻近分类法及模糊分类法进行特征的选择,最后得到Kappa系数0.79的结果,分析原因在于植被、河流因为较好的光谱或几何特征容易被划分,而房屋因为光谱紊乱而没有得到很好的分类。更多还原

【Abstract】 With the popularity of IKNONOS、QuickBird and the appearance of WorldView-1,High-resolution Image has been the main data source in the research of land use/cover. High-resolution image has vast information of space and clear texture. Traditional method of classification can’t extract the useful space information. the classification based on pixel would come to the“pepper-salt effect”, usefulness patch, which is failed in the accuracy assessment. People developed object-oriented classification, which is applied in the land use/cover classification or information extract, from the image segmentation.Object-oriented classification worked base on the object, and define the object by scale, shape/color, compactness/smoothness.Definiens, which is professional software, have some segmentation algorithm, such as quadtree segmentation or chessboard, and the multi-scale is the most popularity. The choosing of the best scale is bother the researchers, many people try to find a rule, such as the most area, as these rule is too complicated, most of the researcher choose the scale by eyes.This paper create the index of ,and choose a set of scale for a hierarchical system of land use and cover, which is including river, road , forest, land ,irrigated land, house, damaged land ,shadow, tree cluster and so on. With the base of segmentation, classified the luc according to the nearest neighbor method and fuzzy classification.With the result of Kappa coefficient is 0.79, maybe, it is classified easily to segmented the vegetation and river, which have a good feature of spectrum or geometry, at the same time, house was wrongly classified with disorder spectrum.更多还原