Evaluation of spectral matching for change detection with CARABAS-II

Change detection using ultra-wideband SAR images processed with a spectral matching filter is investigated. Strip map SAR images from the flight campaign in Vidsel 2002 are used. For two images created with different flight headings the ground reflectivity spectrum is not the same everywhere. We present a spectral matching algorithm that removes components that are not common to both images, which improves change detection performance slightly. Evaluation shows that spectral matching gives slightly better detection performance. With a heading difference of 15°-20°, spectral matching reduced the false alarm ratio from 1/km2 to 0.75/km2 when using an incoherent change detection method with 80% probability of detection. With a coherent change detection method the false alarm ratio changed from 3.5/km2 to 2/km2 with the same probability of detection. It is expected that a larger improvement will be obtained for denser forests.