一种面向栅格的空间-属性双重约束聚类方法

针对栅格数据,传统聚类方法大都基于专题属性进行聚类,分裂了栅格对象的空间特性与专题属性,而兼顾空间与专题属性的现有空间聚类方法又存在算法复杂、参数设置多等问题,因此本文提出了一种面向栅格的空间-属性双重约束聚类算法（A Raster-oriented Clustering Method with Space-Attribute Constraints, RoCMSAC）。RoCMSAC利用栅格数据空间邻域和空间连通特性,重新定义栅格簇的相似性度量准则,通过属性均质簇生成,空间相邻栅格簇合并和空间邻近栅格簇合并3个步骤对栅格数据进行空间-属性双重约束聚类。利用太平洋海域海表温度栅格数据对算法的可行性以及有效性进行验证,并与现有算法进行对比分析。通过实例验证与对比发现：① RoCMSAC方法能够保证栅格簇空间域的邻近性和属性域的均质性;② RoCMSAC方法可发现复杂形状的栅格簇,且算法时间复杂度低,需输入参数较少。

A Raster-Oriented Clustering Method with Space-Attribute Constraints

For dealing with the raster datasets, most of the traditional clustering methods are based on the thematic attribute, which separate the integrities of spatial and thematic characteristics. However, the current clustering methods considering both spatial and thematic characteristics still have great problems such as complicated clusters, computational complexities and many input parameters, etc. Thus, this paper presents a Raster-oriented Clustering Method with Space-Attribute Constraints, named RoCMSAC. The core idea of RoCMSAC uses the spatial contiguities and the connectivity of raster datasets to redefine the similarity measure criterion. The RoCMSAC consists of three steps, i.e. the cluster generation with the homogeneous attributes, the cluster merging with the spatial contiguities and the cluster merging with the spatial vicinities. Finally, the feasibility and effectiveness of the algorithm are validated with the datasets of sea surface temperature in Pacific Ocean. The clusters from RoCMSAC are compared with those from K-Mean and DDBSC. The results show that: (1) RoCMSAC can detect any grid cluster with the complicated shape, which needs less time and fewer input parameters; (2) The clusters from RoCMSAC obtain both the proximity in spatial domain and the homogeneity in attribute one.