A Deep Learning Framework for Joint On-Board InSAR Phase Denoising and Compression

Abstract

In the last decades, Synthetic Aperture Radar and Interferometric Synthetic Aperture Radar instruments have established as very effective and powerful tools in the framework of planetary exploration, featuring the on-board generation of higher-level products. However, in this context the limited down-link capacity has long been a major constraint for SAR and InSAR observations, making the development of efficient on-board compression strategies a critical aspect for such missions. In this work, we present a novel Deep Learning-based approach, through the implementation of a Convolutional AutoEncoder, which allows for the joint denoising and compression of the interferometric phase on board. The proposed network is trained and tested using synthetic datasets, derived starting from real TanDEM-X observations and assuming corresponding InSAR acquisition geometries and underlying topography. Results are assessed with respect to a combination of boxcar filtering and JPEG 2000 compression, which reflects one of the possible strategies reported in the literature. In particular, we focus on three different performance metrics, i.e. denoising capability, data volume reduction and preservation of high-resolution phase details, thus showing the enhanced flexibility of the proposed methodology with respect to a state-of-the-art baseline method.