中文摘要：

传统聚类方法大都是基于空间位置或非空间属性的相似性来进行聚类,分裂了空间要素固有的二重特性,从而导致了许多实际应用中空间聚类结果难以同时满足空间位置毗邻和非空间属性相近。然而,兼顾两者特性的空间聚类方法又存在算法复杂、结果不确定以及不易扩展等问题。为此,本文通过引入直接可达和相连概念,提出了一种基于双重距离的空间聚类方法,并给出了基于双重距离空间聚类的算法,分析了算法的复杂度。通过实验进一步验证了基于双重距离空间聚类算法不仅能发现任意形状的类簇,而且具有很好的抗噪性。

英文摘要：

Most traditional clustering methods only take either the geometric distance or the similarity of attrib utes into account, splitting the dual characteristics of the spatial features. Thus it is difficult for the clustering results in many practical applications to meet the requirement that the clustered features are both nearest in spa tial domain and very similar in attribute domain. So far, some clustering methods which considered dual characteristics of spatial features have many problems, such as algorithm complexity, uncertain clustering results and difficulty for general extension. To solve these problems, this paper proposes a Dual Distance Based Spatial Clustering method （DDBSC）, via utilizing the concepts of dual distance reachability and connection. Meanwhile, the algorithm for the implementation of DDBSC is presented and its complexity is further analyzed. Finally, two experiments demonstrate that the DDBSC algorithm is suitable for arbitrary shape of clusters, and is robust for certain magnitude of noise.