Clustering Algorithm In The Spatial Data Mining

Spatial data mining refers to picking up connotative knowledge, rules or significative modes from spatial database. It has great significance in understanding spatial data and picking up relations between spatial data and non-spatial data. Spatial clustering is a primary field for spatial data mining. It is the task of grouping the objects of a spatial database on some criterias into meaningful subclasses so that the members of a cluster are as similar as possible whereas the members of different clusters differ as much as possible from each other. Spatial clustering can be not only a special technique of spatial data mining but also the technique of pretreatment of other means. Now spatial clustering has many applications in GIS, remote sense, image processing, environment research and so on.At present, many clustering algorithms are available, such as CLARANS, BIRCH, DBSCAN, CLIQUE, and so on. Although many of them have been applied successfully in some fields, many new challenges are emerging. For example, DBSCAN can get the cluster with random shapes, but it has the shortage in the handling of large datasets and high-dimensional datasets. CLIQUE can solve the clustering of high-dimensional datasets, but it cannot resolve the clustering with constraints and the clustering of random shapes, etc. This thesis researches on the clustering faced on obstacles and improving existing algorithms.To deal with the constraints of obstacles, a novel mesh-based clustering algorithm in the presence of obstacles is presented in this thesis, which introduces the concept of obstacle-mesh into the algorithm of CLIQUE. By decomposing obstacles into appropriate meshes, the algorithm can deal with random-shape obstacles. Next the new algorithm is extended in order to resolve the clustering of area entities with random shapes. Moreover the cell-based algorithms have...