Despeckling Sentinel-1 GRD images by deep learning and application to narrow river segmentation

TL;DR

This paper introduces a deep learning-based despeckling method for Sentinel-1 GRD images using a self-supervised framework, significantly enhancing the detection and segmentation of narrow rivers in SAR imagery.

Contribution

The paper develops a novel self-supervised deep neural network for SAR image despeckling, improving structure detection in Sentinel-1 images compared to existing methods.

Findings

Enhanced river segmentation accuracy after despeckling

Robust despeckling against space-variant speckle correlations

Improved detection of narrow river structures

Abstract

This paper presents a despeckling method for Sentinel-1 GRD images based on the recently proposed framework "SAR2SAR": a self-supervised training strategy. Training the deep neural network on collections of Sentinel 1 GRD images leads to a despeckling algorithm that is robust to space-variant spatial correlations of speckle. Despeckled images improve the detection of structures like narrow rivers. We apply a detector based on exogenous information and a linear features detector and show that rivers are better segmented when the processing chain is applied to images pre-processed by our despeckling neural network.