With the fast development of data capturing technologies, voluminous geographical datahas been accumulated in databases and is still increasing rapidly, even the growing velocityis becoming faster. In order to get interesting information, people have to be plunged in tothe sea of data. At the same time our spatial database systems have not provided us efficienttools besides their capabilities in data input, update, basic analysis and display. The usershave stayed in the stage which is drowning in data but starving for knowledge.The voluminous data has posed great challenges to traditional data analysis methods. Datamining or knowledge discovery in databases (KDD) has been emerging as a new researchfield which incorporates the thoughts of machine learning, statistical analysis, neuralnetwork, scientific visualization and data management. Data mining in spatial databasesrefers to the extraction of implicit knowledge, spatial relationships or other useful patternsnot explicitly stored in spatial databases. Spatial databases distinguish from transactiondatabases with complicated data types and relationships among objects. The knowledge thatcan be uncovered to date is on clustering, classifying, association, generalization andevolution. 数据挖掘研究院
This paper puts forward the formal description of spatial clustering and association rulesand analyzes the characteristics of these two kinds of knowledge. First, we implement thehierarchical clustering, which constructs a hierarchical clustering tree based on the distancematrix of objects. A density-based method is proposed, which constructs Delaunaytriangulation and analyzes the edges of the triangles and determines a partitioning length.This method is applied to the scattering points directly and area features with constraint. Inthe pilot work, we take the resident settlement as example and the clustering methodfacilitates the system to determine the classification of street blocks automatically.Successively, an algorithm for constructing constraint Delaunay triangulation is proposed.
The triangulation is used to construct the proximal relationships between street blocks inone cluster. In the topographic map, roads are associated with resident settlement, bridgesetc. The method, which is used to find association rules between them, is primarily based onspatial computation with two steps that employs the object approximation firstly and finerobject itself secondly. Then with the association information, the detailed algorithm to formroad network is described. At the end of this paper the author summarizes the work andproposes the future research directions.【Key words】Data Mining and Knowledge Discovery, Spatial Databases,Spatial Relationships, Spatial Clustering, Spatial Association, ResidentSettlement, Road System, Delaunay Triangulation 数据挖掘研究院
