Deep Learning Methods For Synthetic Aperture Radar Image Despeckling: An Overview Of Trends And Perspectives

Abstract

Synthetic aperture radar (SAR) images are affected by a spatially-correlated and signal-dependent noise called "speckle", which is very severe and may hinder image exploitation. Despeckling is an important task that aims at removing such noise, so as to improve the accuracy of all downstream image processing tasks. The first despeckling methods date back to the 1970's, and several model-based algorithms have been developed in the subsequent years. The field has received growing attention, sparkled by the availability of powerful deep learning models that have yielded excellent performance for inverse problems in image processing. This paper surveys the literature on deep learning methods applied to SAR despeckling, covering both the supervised and the more recent self-supervised approaches. We provide a critical analysis of existing methods with the objective to recognize the most promising research lines, to identify the factors that have limited the success of deep models, and to propose ways forward in an attempt to fully exploit the potential of deep learning for SAR despeckling.

A review of deep-learning techniques for SAR image restoration

“…Extending deep learning methods to polarimetric and/or interferometric SAR data is a hot topic. view on the key elements of deep learning techniques for speckle reduction and invite the interested reader to refer to much more extensive reviews such as [15] and [16].…”

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

“…We look for the shift δ such that the translated profile T δ {p} and its symmetric S{T δ {p}} superimpose at best: δ = arg min δ ||T δ {p} − S{T δ {p}}|| 2 2 . Since the translations of the spectrum that we consider are circular, ||T δ {p}|| 2 2 and ||S{T δ {p}}|| 2 2 are constant for all values of δ. It is then equivalent to estimate δ by maximizing the correlation: δ = arg max δ T δ {p} S{T δ {p}}.…”

“…It is responsible for the strong fluctuations that dramatically degrade the quality of SAR images. Speckle suppression has been the subject of many research works, from the pioneering works of Lee [1] to the most recent techniques based on deep neural networks [2]- [4].…”