一种基于空间邻接关系的k-means聚类改进算法  

A k-means Adapted Algorithm Based on Spatial Contiguity Relations

作　　者：王海起[1] 王劲峰[1] 

机构地区：[1]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室

出　　处：《计算机工程》2006年第21期50-51,75,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(40471111);国家"863"计划基金资助项目(2002AA135230-1);国家"973"计划基金资助项目(2001CB5103)

摘　　要：空间对象不仅具有非空间的属性特征,而且具有与空间位置、拓扑结构相关的空间特征。利用传统的聚类方法对空间对象进行聚类时,由于没有考虑空间关系,同一类的对象可能出现在空间不相邻的位置。基于空间邻接关系的k-means改进算法将相邻对象的空间邻接关系作为约束条件加以考虑,使聚类结果既反映了属性特征的相似程度,又反映了对象的空间相邻状态,从而可以揭示不同类别对象的空间分布格局,因此其比传统的k-means方法更适合于空间对象的聚类分析。Spatial object has not only non-spatial attribute properties but also spatial properties related with space coordinates and topological structures. When using the traditional clustering methods to classify spatial objects, the objects of the stone class may appear in non-adjacent spatial positions because spatial relationships are not been considered. The k-means adapts algorithm based on spatial contiguity relations regards spatial contiguities of the neighboring objects as a restrained condition. So the clustering result not only reflects the similarities of attributes but also reflects spatial＇adjacent relations, and furthermore reviews spatial distribution patterns of different classes. Therefore, this adapted algorithm is more suitable for the clustering analysis of spatial objects than the traditional k-means method.

关 键 词：空间对象 空间邻接关系 邻接矩阵 K-MEANS聚类算法 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]