【作者】 陈春雷；

【导师】 武刚；

【作者基本信息】 北京林业大学 ， 森林经理学， 2011， 博士

【摘要】 由于地理现象的等级层次性和地表的复杂性,带来了遥感研究的空间尺度问题,尺度理论作为地理信息科学的基础理论,也是地理数据的基本性质之一。目前高空间分辨率传感器的发展趋势,造成了遥感信息海量化的现状。林业遥感作为遥感的分支之一,同样有着等级层次的特征,如何融合利用这些冗余的多尺度数据资源,从而提高林业遥感应用中的精度,已经成为全世界林业科技工作者急需解决的问题之一。遥感图像反映的是地学环境的综合性与复杂性特征,而数学方法则是最有力的分析工具。因此,本论文采用数学的方法,根据不同遥感数据源空间分辨率不同的特点,以尺度相关理论为基础,利用多源信息融合理论,挖掘多尺度林业遥感数据的冗余性及互补性,研究多尺度林业遥感数据融合技术的应用问题。基于多源信息融合的优势和制约因素,围绕着尺度的关键问题,本研究以森林资源信息为对象,以像素→特征→决策三个不同的融合级别为主线,研究的主要内容如下：1.在数据预处理阶段,根据不同的数据特点,采用了不同的几何校正模型,比较选择了合理的地形辐射校正方法,从而解决了融合中信息类型的高度相异性和内容的模糊性难题。2.通过一系列的方法,来降低多尺度数据融合带来的复杂性,这些方法包括了波段组合选择、基于景观生态学的廊道思路对研究区域进行区域分割等。3.通过多个评价指标对常用的像素级融合方法进行比较分析,选择了适合的融合方法,据此实现了六种不同空间尺度的多光谱数据源,扩展了光谱空间的覆盖范围。4.针对面向对象分割的评价准则,通过构造平均分割评价指数,得到最优分割尺度参数,从而获得最佳的分割对象集,这不仅丰富了分割评价方法的研究,也将面向像素的传统分析方法转变为面向对象的研究。5.根据尺度研究的关键问题,选择了其中的三个主要内容：最优尺度选择、尺度效应、尺度依赖进行了研究。通过对传统的局部变异和变异函数进行改进,从空间角度为遥感影像的选择提供了客观依据；利用分形几何理论中的分形维数,对遥感的尺度效应提供了研究；基于不同尺度间的对象继承关系,以灰度线性可加性为条件,对不同数据源之间的依赖关系进行了量化研究,并得到了不同尺度针对不同景观区域类型的依赖关系,也使得不同尺度间拓扑关系的研究建立在科学的量化基础上。6.基于尺度依赖的研究成果,通过构造多尺度的对象约束融合模型,在不改变尺度应用抽象程度的基础上,利用知识分类的结果,实现了不同尺度之间的融合,而实验结果也证明了多尺度融合对林业遥感的贡献程度,这也是本研究的主要创新点。更多还原

【Abstract】 Because of the hierarchy of geographical phenomena and complicated landcover types, the issue of spatial scale has been brought to the remote sensing research. As the basis of geographic information science theory, scale theory is also one of the basic properties of geographic data. Now the development of high spatial resolution sensors has brought the status of remote sensing information’s sea quantify. Forestry remote sensing is the branch of remote sensing, and hierarchy is also its character. How to use these redundant data resources and increase the accuracy in forestry remote sensing, which has become one of the issues need to be resolved to forest scientists around the world.Remote sensing images reflect the combining and complicacy of environment in geoscience, in addition mathematical method is the powerful analysis tool which is the main method in the research. According to the different spatial resolution of remote sensing data, as the scale relevantly theory basis, using multi-source information fusion theory, mining the redundancy and complementarity of multi-scale remote sensing data, research the applications of multi-scale remote sensing data fusion technology in forest.Considering the advantages and constraints of multi-source information fusion, concerning the key issues in scale study, as the forest resource information research object, the research line consists of three rank, they are pixel, feture and decision. The paper’s main contents as following:1 In data pre-processing step, different geometric correction models are used for different data types. By comparing results produced by different methods, selecting a reasonable method for radiometric correction of terrain, and this solves the high dissimilarity of information and reduce the ambiguity in contents.2 Through a series of methods to reduce the complexity of study, as selecting different bands combination, based on the corridor ideas of landscape ecology, the study area is partitioned to different regions.3 Through comparing the common resolution fusion methods by multiple evaluation indexes, the adaptive fusion method is obtained. This achieve multi-spectral data sources in six different spatial scales, the coverage of the spectral space is expanded.4 According to the evaluation criteria in object-oriented segmentation, by constructing average segmentation evaluation index, the optimal segmentation scale parameters and the best segmentation object sets are obtained, which not only enriches the segmentation evaluation method, but also change the traditional method for the pixels into object-oriented research.5 According scale key issues research, three main fields are selected in this study such as optimal scale selection, scale effects and scale dependence. Through improving the limitations to adapt to multi-spectral research of local variance and variograms, an objective basis to the choice of remote sensing images from space is provided. By fractal geometry theory and fractal dimension, the remote sensing scaling effect is researched. Based on the inheritance of objects with different scale levels, as the additive of gray linear a condition, a quantitative study about the dependency of different data sources is given, and the dependency relation results are obtained of different scales for different types in diverse landscape areas, while makes the topological relations research during different scales based on a scientific quantitative basis.6 Based on the research results of scale dependence, by constructing multi-scale objects constraints fusion model, without changing the basis of scale abstraction level, with the results of classification on basis of knowledge, the fusion of different scale datas come to true. By the results of experiment, the contribution of multi-scale fusion method to forestry remote sensing is validated, which is the innovation point of the research.更多还原