Data Mining Applications on large RSI Data

摘要

Data Mining becomes more important as more and more data are generated and collected. In this paper, we discuss the application of data mining to Remotely Sensed Imagery (RSI) Data. Since RSI data contains huge amounts of information, it's a high potent area for data mining. Association Rule Mining can be applied to RSI data in which each pixel is a transaction. Discretization needs to be performed first to deal with quantitative data. Equi-length, equisupport and user-defined partition are three ways of discretization. Based on the characteristics of RSI data and the "item group" idea, two pruning techniques are proposed to prune uninteresting rules, thus improving the efficiency. Some pre-processing and post-processing on RSI data can provide a fast way of discovering rules. The application of another two important data mining techniques, clustering and classification, on RSI data, are also described.