Image-to-image-translation to support the interpretation of SAR imagery

Kurzfassung

SAR imagery is nowadays a common source of data in the remote sensing field. Its weather independence working mode and capabilities to measure the 3D profiles of the surfaces are some of its main advantages. However, interpretation of the images produced by this sensor are not easy to interpret as optical imagery. Moreover, the speckle effect produced due to its measurement nature looks like granular noise and degrades the quality of the images. Research conducted during the last years in the area of conditional generative adversarial networks show a feasible approach to process images and learn transfer functions between two different domains. This strategy has been already applied to remote sensing tasks and even to SAR pictures. By using SAR and optical imagery as our input domains, we conduct experiments to generate complementary images that facilitate the interpretation of the original SAR imagery files. Obtained results are not only improved by the remotion of speckle effect, but present optical-like features easier to comprehend by the users.