Feature Based Classification of Interferometric SAR Data for Rapid Mapping

A testbed is proposed for the rapid prototyping of algorithms to extract information from three-dimensional (3D) synthetic aperture radar (SAR) data including the high resolution digital terrain elevation data (DTED), the backscatter image and the phase correlation image. The testbed will operate on a Silicon Graphics workstation with a stereo display and Geographical Information System (GIS) software. The Phase I activity will focus on evaluating algorithms for automated: i) Identification and localization of hydrology drainage networks; ii) Identification of forest canopies; iii) Identification of water bodies and separation from shadow regions; iv) Identification of urban regions and detection of roads, building, bridges and airports. The approach for Phase I will be to first conduct a survey of the existing information extraction algorithms that have been developed for SAR. Using this information and the existing capabilities at Vexcel, we will develop a suite of techniques, the most promising of which will be coded and tested using existing ERIM and JPL IFSAR data. The results will be validated with manually extracted features from optical data. In Phase II an operational system would be developed incorporating a rule based expert system and a commercial GIS. The final system would be designed to produce map quality products form IFSAR data.