The Research On Data Mining System Architecture And Implementary Techniques For Remote Sensing Imagery

With the abundance production and application of Remote Sensing (RS) image, the management of RS data and its processing theory, technique and algorithm needs a breakthrough. People usually need to retrieve different levels of knowledge from very large volume RS data, and then apply this knowledge to RS image segment, so as to improve the efficiency and accuracy of RS image analysis and to establish a RS image based intelligent GIS. This paper put the emphasis on several key techniques of RS image data mining. The main work and structure is as follows.1. In the aspect of RS image data mining system architecture, a RS image data mining processing flowchart based on spatial data warehouse is presented, also on the foundation of this a RS image data mining integrating architecture is presented to implement the whole process of RS image data mining, such as preparing data, building the warehouse, focusing on the data, creating the RSDMM, training the RSDMM and using the RSDMM to predict classifications.2.In the fields of data mining oriented RS image data modeling and feature retrieving, a multi-layer abstract RS image data model for data mining purpose is presented, to provide the rationale for structuring RS image data, building RS image feature database and data warehouse. The thorough view of data model for data mining and schemas of RS image feature database are established.3. As for RS image data warehouse and OLAP, by introducing spatial dimension, spatial index in fact table measures which points to the record set of spatial aggregation, as well as spatial constrains sets, into the traditional MDM, the paper constructs spatial MDM and spatial cuboids. The pointer intersection algorithm is adapted by introduction ORACLE SPATIAL object-relational model and Microsoft data cube construction techniques to vector spatial measure materialization. A RS image data cube is constructed to evaluate effects of each factor on the image classification by OLAP on this cube.4. As for the RS image data mining model (RSDMM) and its implements, The RSDMM's architecture and its creation, training, and prediction is discussed. The design considerations of open data mining algothesis based on MS OLE DB for Data Mining is presented. The integration of ERDAS IMAGINE expert classifier and data mining system is discussed so as to overcome the bottleneck of acquiring knowledge into ERDAS IMAGINE knowledge engineer. Experiments on RSDMM decision tree classification model is presented with the comparisons to ERDAS expert classifier and Hopfield classification model. The experiments show that the RSDMM decision tree classification model takes spatial relationships and other contexts into consideration. It is knowledge driven. The process of knowledge acquire, presentation and application is highly automatic. Its classification accurate is high through the experiments.