项目名称: 基于多源条件随机场的地球物理参数预测模型研究
项目编号: No.61272272
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 韩波
作者单位: 武汉大学
项目金额: 80万元
中文摘要: 本项目将条件随机场引入到多源遥感信息领域,提出基于多源条件随机场的地球物理参数预测模型。该模型的建立过程包括:首先,根据遥感信息的空间分布特点,研究表示空间相关关系的非因果全局概率分布模型,提出遥感信息的空间邻域结构,并验证其相邻相似特性。该结构中节点及节点间连通采用模糊概率表示,支持具有不精确、不全面特点的遥感信息表达,也支持多空间分辨率、多拍摄角度、多质量水平的多源遥感信息融合;其次,通过设计关联势函数、交互势函数、质量信号函数和空间分布信号函数,扩展迭代缩放算法,从而实现多源条件随机场对连续值空间目标的回归模型求解;此外,利用空间邻域结构相邻相似的性质实现空间插值和平滑,从而优化多源条件随机场模型的预测速度。多源条件随机场模型能有效利用多源信息和邻域信息,扩大数据挖掘算法的输入信息量,从而为提高地球物理参数的预测精度和预测速度提供了新的科学方法。
中文关键词: 多源;空间邻域结构;数据挖掘;预测;条件随机场
英文摘要: This project introduces conditional random fields into the field of multi-source remote sensing, and proposes a prediction model for geophysical parameters based on multi-source conditional random fields. The modeling process consists of the following steps: firstly, based on the spatial distribution characteristics of remote sensing, we study the non-causal global probability distribution model for expressing spatial correlations, propose spatial neighborhood structure for remote sensing information, and verify its adjacent similarity features. The nodes and edges in the structure are denoted by fuzzy probability, supporting the expression of inaccurate and incomplete remote sensing information, and also supporting fusion of multi-source remote sensing information with multi-resolution, multi-angle and multi-quality characteristics. Secondly, we design association potential, interaction potential, quality indicator and space distribution indicator, extend the iterative scaling algorithm in order to solve conditional random field regression model for continuous-valued spatial target. In addition, by using the adjacent similarity features of spatial neighborhood structures, we propose spatial interpolation and smoothing algorithms for speeding up the calculation of the multi-source conditional random field model.
英文关键词: multi-source;spatial neighborhood structure;data mining;prediction;conditional random field