Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SAR Despeckling via Regional Denoising Diffusion Probabilistic Model

Xuran Hu, Ziqiang Xu, Zhihan Chen, Zhengpeng Feng, Mingzhe Zhu, LJubisa Stankovic

TL;DR

Speckle noise in SAR images complicates despeckling, especially for large-scale imagery. The authors propose Region Denoising Diffusion Probabilistic Model (R-DDPM), which trains a diffusion process on image regions and employs region-guided inverse sampling to deliver high-quality despeckling across arbitrary scales. Key innovations include regional diffusion for scale flexibility, a lightweight network with DDIM-inspired acceleration, and overlapping-region sampling to prevent edge artifacts. Experiments on Sentinel-1 data show that R-DDPM outperforms state-of-the-art methods in PSNR/SSIM and real-image metrics (ENL, EPI), with low GPU memory footprint and straightforward deployment.

Abstract

Speckle noise poses a significant challenge in maintaining the quality of synthetic aperture radar (SAR) images, so SAR despeckling techniques have drawn increasing attention. Despite the tremendous advancements of deep learning in fixed-scale SAR image despeckling, these methods still struggle to deal with large-scale SAR images. To address this problem, this paper introduces a novel despeckling approach termed Region Denoising Diffusion Probabilistic Model (R-DDPM) based on generative models. R-DDPM enables versatile despeckling of SAR images across various scales, accomplished within a single training session. Moreover, The artifacts in the fused SAR images can be avoided effectively with the utilization of region-guided inverse sampling. Experiments of our proposed R-DDPM on Sentinel-1 data demonstrates superior performance to existing methods.

SAR Despeckling via Regional Denoising Diffusion Probabilistic Model

TL;DR

Speckle noise in SAR images complicates despeckling, especially for large-scale imagery. The authors propose Region Denoising Diffusion Probabilistic Model (R-DDPM), which trains a diffusion process on image regions and employs region-guided inverse sampling to deliver high-quality despeckling across arbitrary scales. Key innovations include regional diffusion for scale flexibility, a lightweight network with DDIM-inspired acceleration, and overlapping-region sampling to prevent edge artifacts. Experiments on Sentinel-1 data show that R-DDPM outperforms state-of-the-art methods in PSNR/SSIM and real-image metrics (ENL, EPI), with low GPU memory footprint and straightforward deployment.

Abstract

Speckle noise poses a significant challenge in maintaining the quality of synthetic aperture radar (SAR) images, so SAR despeckling techniques have drawn increasing attention. Despite the tremendous advancements of deep learning in fixed-scale SAR image despeckling, these methods still struggle to deal with large-scale SAR images. To address this problem, this paper introduces a novel despeckling approach termed Region Denoising Diffusion Probabilistic Model (R-DDPM) based on generative models. R-DDPM enables versatile despeckling of SAR images across various scales, accomplished within a single training session. Moreover, The artifacts in the fused SAR images can be avoided effectively with the utilization of region-guided inverse sampling. Experiments of our proposed R-DDPM on Sentinel-1 data demonstrates superior performance to existing methods.
Paper Structure (9 sections, 9 equations, 5 figures, 2 tables)

This paper contains 9 sections, 9 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dinoising Diffusion Probabilistic Model
  4. Conditional Diffusion
  5. Regional Restoration
  6. Experimental Results
  7. Experiment Settings
  8. Comparison with State-of-the-Art
  9. Conclusion

Figures (5)

  • Figure 1: An overview of the diffusion and reverse processes of conditional diffusion model.
  • Figure 2: Illustration of regional restoration: diffusion sampling involves moving the 'window' $x_t^k$ across diverse regions, culminating in the final despeckled outcome by averaging overlapping regions.
  • Figure 3: The demonstration of experiments on large-scale images. The first row showcases the despeckling reconstruction results, the second row displays the residual results.
  • Figure 4: The demonstration of experiments on small-scale images. The first row showcases the despeckling reconstruction results, the second row displays the residual results.
  • Figure 5: The demonstration of experiments on real SAR images, where the first row showcases the despeckling reconstruction results, the second row displays the residual results.