As if by magic: self-supervised training of deep despeckling networks with MERLIN

TL;DR

This paper introduces MERLIN, a self-supervised method for training deep neural networks to despeckle SAR images without needing speckle-free references, enabling large-scale and hassle-free training.

Contribution

The paper presents MERLIN, a novel self-supervised training strategy for deep despeckling networks that only requires a single SAR image and accounts for sensor-specific spatial correlations.

Findings

MERLIN effectively trains despeckling networks without ground truth images.

Networks trained with MERLIN outperform traditional supervised methods.

The approach enables large-scale, hassle-free training using large image archives.

Abstract

Speckle fluctuations seriously limit the interpretability of synthetic aperture radar (SAR) images. Speckle reduction has thus been the subject of numerous works spanning at least four decades. Techniques based on deep neural networks have recently achieved a new level of performance in terms of SAR image restoration quality. Beyond the design of suitable network architectures or the selection of adequate loss functions, the construction of training sets is of uttermost importance. So far, most approaches have considered a supervised training strategy: the networks are trained to produce outputs as close as possible to speckle-free reference images. Speckle-free images are generally not available, which requires resorting to natural or optical images or the selection of stable areas in long time series to circumvent the lack of ground truth. Self-supervision, on the other hand, avoids the use of speckle-free images. We introduce a self-supervised strategy based on the separation of the real and imaginary parts of single-look complex SAR images, called MERLIN (coMplex sElf-supeRvised despeckLINg), and show that it offers a straightforward way to train all kinds of deep despeckling networks. Networks trained with MERLIN take into account the spatial correlations due to the SAR transfer function specific to a given sensor and imaging mode. By requiring only a single image, and possibly exploiting large archives, MERLIN opens the door to hassle-free as well as large-scale training of despeckling networks. The code of the trained models is made freely available at https://gitlab.telecom-paris.fr/RING/MERLIN.