基于感知编组规则的遥感影像分析马尔柯夫随机场图模型研究

A Markov Random Field Graph Model (MRFG) for remote sensing image analysis based on perceptual organization rules is introduced in this project, aiming at solving the problem of multi-feature comprehensive analysis of image segments. This problem is caused by the irregular shape of image segments and the difficulty of spatial feature quantitative analysis during object-based high resolution remote sensing image interpretation. Spatial structure features of image segments were analyzed quantitatively and mathematical formulation were given according to perceptual organization rules, and together with other spectral features of image segments the feature space is established. MRFG model is then constructed on the basis of random field and graph theory, used for multi-feature image segment analysis. Meanwhile energy function of MRFG and related model solution method were designed according to the different forms of the features. Land use/cover change detection is selected as a kind of practical applications for experimental verifying and updating the model, so as to provide a scientific method as well as a theoretical foundation in the field of Geographical National Condition Monitoring.

本课题提出了一种基于感知编组规则的遥感影像分析马尔柯夫随机场图模型（MRFG），用以解决高分辨率遥感影像解译中由于像斑外形不规则、空间特征定量化表达困难造成的像斑多特征难以综合分析的问题。研究以感知编组为基础完成空间结构性特征的定量化分析和数学表达，结合已有的光谱特征共同建立特征空间；在此基础上以随机场理论和图论为基础建立马尔柯夫随机场图模型，用以进行像斑的多特征综合分析；同时针对光谱、空间特征表达方式不同的特点设计能量函数并完成模型解算；以土地利用变化检测为实际应用验证和更新模型，从而为地理国情监测领域提供科学方法和理论依据。

本项目收集研究了国内外关于基于像斑的遥感影像处理、空间关系分析以及马尔柯夫随机场三方面的研究进展，以土地覆盖/利用和建筑物提取为研究对象，主要进行了像斑空间结构特征的提取与应用，以及马尔柯夫随机场图模型构建方面的研究。项目首先进行了像斑光谱、空间、结构等多类型特征的提取研究，在此基础上进行了基于多特征的建筑物像斑提取方法设计和实验；以土地覆盖/利用分类为研究目的，利用Apriori算法建立了土地覆盖/利用像斑类别空间关联规则，并通过实验评估了空间关联规则在基于像斑的土地覆盖/利用分类中的作用；构建了适用于建筑物像斑分析的马尔柯夫随机场图模型，定义了多层次像斑分析的随机场邻域，并用非参数的方式对模型进行了求解，完成了建筑物像斑的分析和识别。项目组还根据实验需求建立了基于VC++2010和ArcEngine10.1的影像像斑处理平台。