Complex-Valued Autoencoders with Coherence Preservation for SAR

Conference: EUSAR 2022 - 14th European Conference on Synthetic Aperture Radar
07/25/2022 - 07/27/2022 at Leipzig, Germany

Proceedings: EUSAR 2022

Pages: 6Language: englishTyp: PDF

Authors:
Mohammadi Asiyabi, Reza; Anghel, Andrei (Research Center for Spatial Information (CEOSpaceTech), University Politehnica of Bucharest, Romania)
Datcu, Mihai (Research Center for Spatial Information (CEOSpaceTech), University Politehnica of Bucharest, Romania & Earth Observation Center (EOC), German Aerospace Center (DLR), Wessling, Germany)
Nies, Holger (Center for Sensor Systems (ZESS), University of Siegen, Germany)

Abstract:
Deep learning techniques have attracted many interests in various fields, including Synthetic Aperture Radar (SAR). A few researches have extended the deep learning framework into the complex-domain to exploit the phase and amplitude of the Complex-Valued (CV)-SAR data. In this study, a complex-valued Convolutional AutoEncoder (CV-CAE) is developed for CV-SAR data reconstruction from the lower resolution azimuth subaperture images. Finally, the performance of the CV-CAE is evaluated, in terms of the reconstruction quality and the coherency preservation, and showed that the CV-CAE is capable of reconstructing the CV-SAR images and preserving the coherency, with a remarkably small training dataset.