层次聚类算法在空间数据分类上的研究及仿真

Research and Simulation of Spatial Data Classification Using Hierarchical Clustering

当前众多聚类算法忽略了对样本的噪点数据处理,影响了分类结果。为提高聚类算法对空间数据的处理能力,减少噪点影响,对聚类算法中的凝聚嵌套算法(AGNES)进行研究与分析,提出一种带降噪预处理的AGNES,算法使聚类后每一个类的元素个数满足用户预先定义的标准,将改进后的算法应用于空间数据分类,可以有效解决边缘"噪点"对全局分类干涉严重的问题.最后,利用VC++语言和基于COM的MapOb jects组件技术实现了基于上述改进算法的仿真软件C ity-C ls,以验证算法的可行性与有效性,并从实验结果中得出了一些有益的结论。

Most of the clustering algorithms ignore dealing with the noise data of the samples, which influence the classification. For the purpose of improving the capability of the clustering algorithms for dealing with the spatial data, the paper analyzes the Agglomerative Nesting （AGNES） which is a clustering algorithm. The preprocessed AGNES based on noise immunity was proposed to make the number of elements in each classes after clustering accordant to the criterion defined by users. The improved algorithm was used to resolve efficiently the problem that the edge ＂noise＂ data have a strong impact on the whole classification. At last, an application CityCls was implemented to validate the algorithm by using the MapObjects components based on COM in the Visual C ＋ ＋ environment, and some useful conclusions were made.