Transformer-based SAR Image Despeckling

TL;DR

This paper presents a transformer-based neural network for despeckling SAR images, leveraging global dependencies to improve image quality, trained on synthetic data, and outperforming traditional and CNN methods.

Contribution

The paper introduces a novel transformer-based architecture specifically designed for SAR image despeckling, capturing global image dependencies for enhanced performance.

Findings

Significant improvement over traditional despeckling methods.

Outperforms CNN-based approaches on synthetic and real SAR images.

Effective end-to-end training with synthetic speckled data.

Abstract

Synthetic Aperture Radar (SAR) images are usually degraded by a multiplicative noise known as speckle which makes processing and interpretation of SAR images difficult. In this paper, we introduce a transformer-based network for SAR image despeckling. The proposed despeckling network comprises of a transformer-based encoder which allows the network to learn global dependencies between different image regions - aiding in better despeckling. The network is trained end-to-end with synthetically generated speckled images using a composite loss function. Experiments show that the proposed method achieves significant improvements over traditional and convolutional neural network-based despeckling methods on both synthetic and real SAR images.