有限混合密度模型及遥感影像EM聚类算法

遥感信息是地球表层信息的综合反映，由于地球表层系统的复杂性和开放性，地表信息是多维的、无限的、遥感信息传递过程中的局限性以及遥感信息之间的复杂相关性，决定了遥感信息其结果的不确定性和多解性，遥感信息具有一定的统计特性，同时又具有高度的随机性和复杂性，在特征空间中往往表现为混合密度分布，针对遥感信息这种统计分布的复杂性，提出了有限混合密度的期望最大（EM）分解模型，该模型假设总体分布可分解为有限个参数化的密度分布，通过EM迭代计算可估计出各密度分布的最大似然参数集；将有限混合EM聚类算法应用于遥感影像的聚类分析中，并与传统统计聚类方法进行了比较，比较结果表明，其对复杂地物的区分具有优势，另外在融合专家知识、初始化等方面具有扩展能力。

Finite Mixture Model and Its EM Clustering Algorithm for Remote Sensing Data

Generally, the analyzed results from remote sensing data are uncertain and multi solution, which is determined by the characteristics of global surface information being multi dimensional and infinite. Therefore, remote sensing information has some degree of definite statistical characteristic, but as well as holds the high randomness and complexity, which generally behaves as mixture density distribution in feature space. In allusion to the complexity of statistical distribution of remote sensing information, in this study we firstly introduce into the finite mixture model and its expectation maximization(EM) algorithm for decomposing the mixture distribution into finite parametric density distributions in order to simulate or approach the whole mixture distribution. By the model it should be firstly assumed that whole distribution could be separated into infinite parametric density distributions, then by EM iterative computation the maximum likelihood parameters of each proportional distribution can be estimated. Furthermore, the finite mixture model and its EM algorithm are extended to clustering algorithm for remotely sensed data. By the experimental case, the EM clustering algorithm is synthetically compared with conventional statistical clustering algorithm. The results show that the EM algorithm has several particular advantages such as self adaptive decision for clustering number, extensibility of prior knowledge integration and free initialization, etc.