KES: knowledge enabled services for better EO information use

Abstract

Modem imaging sensors, especially those aboard satellites, continuously deliver enormous amounts of data. These represent typical cases, where users need automated tools to discover, explore and understand the contents of large image databases. The state-of-the-art image catalogues allow only queries based on data like geographical position, acquisition date, sensor type, etc. and not on image content. Once the search data have been defined, these catalogues show the associated image to the user, leaving the interpretation task to his specific knowledge. The call up a synergy between stochastic modeling, knowledge discovery, semantic representation, up to build a collaborative environment permitting to share the knowledge between heterogeneous user communities. In a previous project (Knowledge Driven Information Mining in Remote Sensing Image Archives - KIM) methods to associate concepts to images were developed for the ESA (European Space Agency). The consortium was composed by ACS (Advanced Computer Systems. Rome). DLR (German Aerospace Center) and ETH (Swiss Federal Institute of Technology. Zurich). This result has been achieved by extracting primitive image features (i.e.: texture, geometrical shapes, spectral information), and by providing the user with a simple, graphic interface to define weighted combinations of these features and to associate concepts (labels) to them through positive and negative examples. The weighted combinations of primitive image features and the associated semantic labels can then be applied to the entire dataset. and not only to the image from which they were defined