COMPARISON OF CONVOLUTIONAL NEURAL NETWORKS FOR CLOUDY OPTICAL IMAGES RECONSTRUCTION FROM SINGLE OR MULTITEMPORAL JOINT SAR AND OPTICAL IMAGES

Abstract. With the increasing availability of optical and synthetic aperture radar (SAR) images thanks to Sentinel constellation, explosion deep learning, new methods have emerged in recent years tackle reconstruction that are impacted by clouds. In this paper, we focus on evaluation convolutional neural networks use jointly SAR retrieve missing contents one single polluted image. We propose a simple framework ease creation datasets for training nets targeting image reconstruction, validation machine learning based or deterministic approaches. These quite different terms input constraints, comparing them is problematic task not addressed literature. show how space partitioning data structures help query samples cloud coverage, relative acquisition date, pixel validity proximity between images. generate several compare reconstructed from pair image, versus multiple pairs, traditional approach performing interpolation temporal domain.