项目名称: 遥感分类中的尺度效应机理与多尺度遥感数据分类方法研究
项目编号: No.40871161
项目类型: 面上项目
立项/批准年度: 2009
项目学科: 金属学与金属工艺
项目作者: 柏延臣
作者单位: 北京师范大学
项目金额: 40万元
中文摘要: 多分辨率的遥感数据在为使用者提供了更多数据选择的同时,也提出了不同分辨率遥感数据专题分类中的尺度效应问题、尺度选择问题和综合多尺度遥感数据提高分类精度的问题。本项目针对遥感数据分类中尺度效应的机理,通过模拟研究建立了具有代表性的遥感分类尺度效应机理模型;针对粗分辨率条件下亚像元尺度空间格局对尺度效应的影响,利用逻辑回归建立了同时考虑像元和亚像元尺度空间格局的尺度效应模型;针对遥感数据分类中的最有尺度选择问题,发展了基于信息熵的最有尺度选择方法;在综合多尺度遥感数据进行分类的方法方面,提出了基于单一尺度数据分类不确定性信息,自下而上集成决策的多尺度分类算法,并针对多尺度分类中数据分辨率的选择问题,提出了在对高分辨率数据进行分割统计基础上优化选择多尺度数据,再进行多尺度分类的策略;在景观格局对遥感数据分类尺度效应影响研究的基础上,提出了基于景观格局的超分辨率制图方法。通过和国外已有的经典方法对比,检验了本方法的有效性,并将其用于真实遥感数据的超分辨率制图。
中文关键词: 尺度效应;景观结构;多尺度分类;不确定性;超分辨率制图
英文摘要: The scale is one of the fundmental issues in remote sensing data classification. This project focus on modeling the mechanisms of scale effect in remote sensing data classification, the method to selecting appropriate resolution for optimal classification, and the method to integrating multi-resolution remeote sensing data for land cover classification. we developed a landscape pattern based model by using the simulated multi-pattern remote sensing data and the multivariable stepwise regression analysis. To incoporate the spatial structure at both pixel scale and sub-pixel scale,we developed a logistic regression model that can predict the classification uncertainty at any spatial resolution. A entropy based method was developed to select the appropriate spatial resolution of remote sensing data for land cover classification. As for the multi-resolution remote sensing data classification method, we developed a uncertainty based method which can integrating multiple classification at any scale based on their uncertainties by the top-down integrating way. For optimally select the resolutions of remote sensing data for multi-resolution remote sensing data classification, we developed a strategy that determining the candidate resolutions by segment the high resolution data and explore the size distribution of the segments. In addition, we developed landscape pattern based super-resolution mapping method. Comparison with existing methods revealed the superority of our method in both 2-class and multi-class super-resolution mapping.
英文关键词: Scale effect; Landscape Structure; Multi-resolution Classification; Uncertainty; super-resolution mapping