Geometry-Aware Deep Learning for InSAR Data Synthesis

Abstract:
Many applications in the field of Earth Observation can benefit from simulated remote sensing images. With the possibilities of deep learning evolving, research on the synthesis of artificial remote sensing images by means of deep neural networks is progressing as well. Generating synthetic aperture radar (SAR) data is still largely limited to intensity images as processing of complex-valued numbers by conventional neural networks poses challenges. With this work, we propose a basic architecture to circumvent these obstacles. Moreover, we demonstrate that the synthesis quality of artificial SAR interferograms can be improved by constraining the network with information from observation geometry.