Improved Arithmetic Of High Dimensional Spatial Data Clustering And The Applications In Informatization Of Agriculture

With the coming of the information age, lots of geographical data with various properties have been collected and applied. This makes the mining of high dimensional spatial data a very important domain. The work of this paper is a part of nation 863 information and technology project, "the Research and Development of environment integrated with intelligent systems", which is taken by the lab of knowledge ware and knowledge engineering of Jilin University. The main aim of the whole study is the research of the platform which can support the web intelligent application systems. The platform is composed of case middleware, knowledge access middleware, spatio-temporal data middleware, communication middleware and data access middleware. The spatial data mining module which this paper researched is a part of spatio-temporal data middleware. The main achievement is to bring out ENSCM (the Extending Neighbored Space Clustering Method). Using this method we can do the spatial data mining, which can be used in precision agriculture. At the same time we have done the agriculture WebGIS, which is based on MapXtreme, using the technique of middleware. In this system we developed agriculture breed partition middleware, made precision agriculture can do well in our country. First we systemic introduced the methods and tools of data mining, including data characterization, data discrimination, association analysis, classification, prediction and clustering. We specially described clustering analysis, including partitioning methods, hierarchical methods, density-based methods and grid-based methods. ENSCM is a density-based and grid-based method which can do clustering on subspace of high dimensional spatial data. Because traditional method failed to consider the affect of neighbored grids, the results of clustering have many problems, e.g. unsmoothed clustering and wrong judgement of clustering boundary. ENSCM adds the affect of neighbored grids when doing clustering. After doing experiments we decided that the considered size is 1/2 extension of grid. In this extended area we make certain the weight of each data point. At the same time we expand the traditional clustering method, make the extension from four directions to including gradient directions. At the end of clustering we also do the detections of outlier and clustering boundary. From the results of experiments, we found that ENSCM conquers the unsmoothed clustering and it can do well at the clustering boundary.