Exploiting different training approaches for CNN based solutions in SAR image despeckling

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

Proceedings: EUSAR 2022

Pages: 4Language: englishTyp: PDF

Authors:
Vitale, Sergio; Ferraioli, Giampaolo (Dipartimento di Scienze e Tecnologie, Università Parthenope di Napoli, Italy)
Pascazio, Vito (Dipartimento di Ingegneria, Università Parthenope di Napoli, Italy)

Abstract:
Despite the effort spent in the last forty years for SAR image despeckling, the problem is still far from being solved. In the last years, the important success of deep learning methods has run over the SAR image despeckling, too. The lack of a ground truth and the difficult characterization of the SAR image make the application of deep learning very challenging. A study about the effect of different training dataset is fundamental for a correct application of such methods. In this paper a deeper analysis of the outcomes of a recent study on different training approaches is carried out.