Hybrid approach for georeferencing RadarSat2 images

Target geolocation in radar images is becoming more crucial with the advent of high-resolution sensors and the variety of acquisition modes. Each image acquisition system produces specific geometric distortions in its raw images and consequently the geometry of these images does not correspond to the terrain or to a specific map projection of end-users. The geometric distortions vary according to different factors such as the acquisition system (the platform, the sensor and other measuring instruments), as well as the observed surface (problem of earth model, effects of its rotation and the problems inherent to the relief [8]). Nevertheless, it is possible to make general categorizations of these distortions. For RADARSAT 2, image acquisition can be achieved in a wide range of viewing directions [6], and at different resolutions, which implies different geometric distortions, specific to each type of acquisition mode. We also have to take into account the range nonlinearities caused by both the height of target regions and the side-looking acquisition mode of SAR images. We propose a georeferencing process for RADARSAT 2 images, using image metadata. Three levels of treatment were achieved: extracting and restructuring the orbital data from the header file; global modelling of the random distortions to achieve georeferencing, and finally rectifying the range position error caused by elevation in the slant range plane. The process was tested on a set of image data acquired by the RADARSAT2 satellite in both quadri and bipolarisation, covering an area of the capital "Algiers" (Algeria).