基于尺度空间的分层聚类方法及其在遥感影像分类中的应用

Scale Space Based Hierarchical Clustering Method and its Application to Remotely Sensed Data Classification

基于尺度空间的分层聚类方法（ＳＳＨＣ）是一种以热力学非线性动力机制为理论基础的新型聚类算法，是视觉松弛化过程的模拟。与传统基于统计方法的聚类算法相比较，ＳＳＨＣ具有样本空间可服从自由分布、通过规则可获取最优聚类中心点及类别、可在聚类过程中融合后验知识等优点。本文从聚类系统的热力学运动机制和视觉模拟过程出发，对ＳＳＨＣ聚类算法进行综合分析，并对如何生成聚类树的过程进行详细描述，提出了通过融合点的部分自由能（ＦＦＥ）和所属聚类子树所包含叶结点最小点集等规则来获取对最优聚类中心点决策；最后对ＳＳＨＣ算法在多波段遥感影像分类中的应用模型进行详细探讨，结论认为相对于传统的统计聚类方法，ＳＳＨＣ聚类算法具有更多的灵活性和实用性。

In Pattern recognition and image processing, the major application areas of cluster analysis, human eyes seem to posses a singular aptitute to group objects and find important structures in an efficient and effective way. Scale Space Based Hierarchical Clustering Method (SSHC) posses a nonlinear dynamical mechanism which simulates the human vision system in relaxing process to a object from legible extend to all blurring extend. The main advantages of SSHC method are: (1) it's distribution of feature space could be assumed free, (2) by FFE decision rules the finest clustering centers and number can be easily extracted, (3) the outside knowledge can be integrated with the process of clustering fusion. In this article,the SSHC's thermodynamic and simulating vision mechanism is analyzed at first, then the SSHC algorithm and the procedure of making the clustering tree are described in detail, in which the decision rule how to acquire the finest clustering centers from clustering tree by Fractional Free Energy (FFE) of each fusion point and the points number of leaf nodes under which the son-tree owns is presented out. At last, we present out the framework of SSHC based on a case of Remotely Sensed Data Classification. We attain such a result that SSHC based RS classification method holds more practicality and flexibility than others RS Clustering Methods.