Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-Supervised Score-Based Despeckling for SAR Imagery via Log-Domain Transformation

Junhyuk Heo

TL;DR

This paper tackles SAR despeckling by addressing multiplicative Gamma speckle with a self-supervised, score-based approach operating in a log-domain. It combines a Log-Yeo-Johnson transform to Gaussianize residual noise with a Corruption2Self–inspired objective, enabling learning of the score function without clean ground truth. The proposed S^4DM method demonstrates improved speckle suppression and detail preservation, with competitive or faster inference compared to existing baselines, validated on Sentinel-1 data using ENL as a quantitative metric. This approach offers practical, scalable despeckling for SAR imagery in real-world settings.

Abstract

The speckle noise inherent in Synthetic Aperture Radar (SAR) imagery significantly degrades image quality and complicates subsequent analysis. Given that SAR speckle is multiplicative and Gamma-distributed, effectively despeckling SAR imagery remains challenging. This paper introduces a novel self-supervised framework for SAR image despeckling based on score-based generative models operating in the transformed log domain. We first transform the data into the log-domain and then convert the speckle noise residuals into an approximately additive Gaussian distribution. This step enables the application of score-based models, which are trained in the transformed domain using a self-supervised objective. This objective allows our model to learn the clean underlying signal by training on further corrupted versions of the input data itself. Consequently, our method exhibits significantly shorter inference times compared to many existing self-supervised techniques, offering a robust and practical solution for SAR image restoration.

Self-Supervised Score-Based Despeckling for SAR Imagery via Log-Domain Transformation

TL;DR

This paper tackles SAR despeckling by addressing multiplicative Gamma speckle with a self-supervised, score-based approach operating in a log-domain. It combines a Log-Yeo-Johnson transform to Gaussianize residual noise with a Corruption2Self–inspired objective, enabling learning of the score function without clean ground truth. The proposed S^4DM method demonstrates improved speckle suppression and detail preservation, with competitive or faster inference compared to existing baselines, validated on Sentinel-1 data using ENL as a quantitative metric. This approach offers practical, scalable despeckling for SAR imagery in real-world settings.

Abstract

The speckle noise inherent in Synthetic Aperture Radar (SAR) imagery significantly degrades image quality and complicates subsequent analysis. Given that SAR speckle is multiplicative and Gamma-distributed, effectively despeckling SAR imagery remains challenging. This paper introduces a novel self-supervised framework for SAR image despeckling based on score-based generative models operating in the transformed log domain. We first transform the data into the log-domain and then convert the speckle noise residuals into an approximately additive Gaussian distribution. This step enables the application of score-based models, which are trained in the transformed domain using a self-supervised objective. This objective allows our model to learn the clean underlying signal by training on further corrupted versions of the input data itself. Consequently, our method exhibits significantly shorter inference times compared to many existing self-supervised techniques, offering a robust and practical solution for SAR image restoration.
Paper Structure (26 sections, 12 equations, 3 figures, 1 table)

This paper contains 26 sections, 12 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Score-Based Generative Models
  4. Method
  5. Denoising Principle and Score Estimation
  6. Speckle Normalization via LYJ Transform
  7. Self-Supervision via Data Corruption
  8. Gaussian Diffusion Step.
  9. Two Tweedie Expectations.
  10. Formulating the Self-Supervised Objective.
  11. Training Objective
  12. Inference
  13. Experiments
  14. Experimental Setup
  15. Dataset and Preprocessing.
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Overall framework of the proposed self-supervised SAR despeckling method ($\text{S}^4\text{DM}$). The diagram illustrates the transformation to the Log-Yeo-Johnson domain, the self-supervised training loop involving data corruption and reconstruction, and the inference path to obtain the despeckled image $X_{target}$.
  • Figure 2: Despeckling results for Agricultural area Image 1, from Table \ref{['tab:quantitative_results']}. (a) Sentinel-2 optical reference. (b) Sentinel-1 input. (c)-(g) Outputs from FANS, SAR-BM3D, DIP, S3DIP, and $\text{S}^4\text{DM}$ (Ours). See Appendix \ref{['app:additional_qualitative_results']} for results on all benchmark dataests.
  • Figure 3: Additional qualitative despeckling results for test images (Image 2, Image 3, and Image 4), supplementing the quantitative metrics presented in Table \ref{['tab:quantitative_results']}. Subfigures (a)-(g) display the results for Agricultural area Image 2. Subfigures (h)-(n) display the results for Mountain area Image 3. Subfigures (o)-(u) display the results for Mountain area Image 4. For each image, the sequence shown is: (a) Sentinel-1 Input, followed by the outputs from (b) FANS cozzolino2013fast, (c) SAR-BM3D parrilli2011nonlocal, (d) DIP ulyanov2018deep, (e) S3DIP albisani2025self, (f) S3DIP+lf albisani2025self, and (g) $\text{S}^4\text{DM}$ (Ours).