研究可分解马尔科夫网(decomposab le M arkov network DMN)的概念、方法;分析它在空间数据挖掘中的作用与意义。与以往关于DMN的研究不同,本文直接将DMN的结构作为推理依据或应用于问题求解,扩大DMN概念和方法的应用范围。以多光谱遥感为例广泛研究以多种迹度量建立多波段遥感图像间的马尔科夫网,用以解释波段组合效果;以视频图像为例广泛研究以多种粒度(节点数)建立视频图像间的马尔科夫网,通过网络结构分析检测视频图像中的目标差异,用以定位和跟踪违章车辆。研究表明马尔科夫网可以很好地揭示空间数据间的抽象近邻关系,并且这种网络自身就具有表达知识的意义。

The paper studies concept and approach of decomposable Markov network(DMN).Unlike previous studies on DMN,our method directly employs the construct of DMN as the evidence of inference or for problem solving,and enlarges application range of DMN's concept and approach.It analyses DMN's role and importance in spatial data mining,and deeply investigates several score metrics used in constructing DMN between multi-bands remote sensing images,so as to interpret the result of fusion in them,and realizes optimal fusion of bands.Another example of the factual video images with traffic rule violation is also investigated by novel concept of DMN's graininess.Several graininess(nodes) are used to construct DMN between video images for detecting abnormality in them.Hence, the vehicle of traffic peccancy is located and traced.The result of video image simulation shown that our method is feasible and effective.The researches indicate that the DMN may reveal abstract adjacent relations existed in spatial data;and the network itself has capabilities of showing knowledge.