An object-based information extraction methodology incorporating a-priori spatial information

The progress in image data mining over the last years is significant but barely known in the GIScience community. Conversely, data mining methods rarely make use of existing spatial information. This paper describes a methodology to extract particular knowledge from spatial data. First step is to define a generic rule set similar to a supervised classification. Metainformation such as the bounding box of an image or the image centroid is straightforwardly utilized to automatically derive information from Spatial Data Infrastructures (SDI). The rule sets are then applied to other images taken by the same sensor through automated adjustments according to the metadata. We demonstrate the degree of automation for two ASTER images, one from Kashmir and one from Zimbabwe based on a worldwide data set of biogeographic regions. This successful blind test illustrates the potential to directly utilizing SDIs within a remote sensing data classification process.