Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ground-based image deconvolution with Swin Transformer UNet

Utsav Akhaury, Pascale Jablonka, Jean-Luc Starck, Frédéric Courbin

TL;DR

This work tackles fast, high-resolution deconvolution of ground-based astronomical images by framing the restoration as an ill-posed inverse problem and introducing a Swin Transformer UNet (SUNet) within a two-step pipeline that starts with Tikhonov deconvolution. A novel debiasing step based on multi-resolution support is added to mitigate flux bias in neural-network outputs, enhancing scientific reliability for galaxy analyses. Compared with classical methods like Firedec, SUNet achieves superior resolution recovery, robust generalisation to diverse noise properties, and orders-of-magnitude faster processing ($\approx$15 ms per image), enabling scalable analysis of large survey datasets. Applied to EDisCS galaxies with multi-band FORS2 data, the approach supports detailed clump detection and reveals a link between disc colour and clump abundance, illustrating the method’s potential for studying structure formation in the distant universe from ground-based imagery. The work provides publicly available code and trained models to foster reproducibility and broad adoption in observational cosmology.

Abstract

As ground-based all-sky astronomical surveys will gather millions of images in the coming years, a critical requirement emerges for the development of fast deconvolution algorithms capable of efficiently improving the spatial resolution of these images. By successfully recovering clean and high-resolution images from these surveys, the objective is to deepen the understanding of galaxy formation and evolution through accurate photometric measurements. We introduce a two-step deconvolution framework using a Swin Transformer architecture. Our study reveals that the deep learning-based solution introduces a bias, constraining the scope of scientific analysis. To address this limitation, we propose a novel third step relying on the active coefficients in the sparsity wavelet framework. We conducted a performance comparison between our deep learning-based method and Firedec, a classical deconvolution algorithm, based on an analysis of a subset of the EDisCS cluster samples. We demonstrate the advantage of our method in terms of resolution recovery, generalisation to different noise properties, and computational efficiency. The analysis of this cluster sample not only allowed us to assess the efficiency of our method, but it also enabled us to quantify the number of clumps within these galaxies in relation to their disc colour. This robust technique that we propose holds promise for identifying structures in the distant universe through ground-based images.

Ground-based image deconvolution with Swin Transformer UNet

TL;DR

This work tackles fast, high-resolution deconvolution of ground-based astronomical images by framing the restoration as an ill-posed inverse problem and introducing a Swin Transformer UNet (SUNet) within a two-step pipeline that starts with Tikhonov deconvolution. A novel debiasing step based on multi-resolution support is added to mitigate flux bias in neural-network outputs, enhancing scientific reliability for galaxy analyses. Compared with classical methods like Firedec, SUNet achieves superior resolution recovery, robust generalisation to diverse noise properties, and orders-of-magnitude faster processing (15 ms per image), enabling scalable analysis of large survey datasets. Applied to EDisCS galaxies with multi-band FORS2 data, the approach supports detailed clump detection and reveals a link between disc colour and clump abundance, illustrating the method’s potential for studying structure formation in the distant universe from ground-based imagery. The work provides publicly available code and trained models to foster reproducibility and broad adoption in observational cosmology.

Abstract

As ground-based all-sky astronomical surveys will gather millions of images in the coming years, a critical requirement emerges for the development of fast deconvolution algorithms capable of efficiently improving the spatial resolution of these images. By successfully recovering clean and high-resolution images from these surveys, the objective is to deepen the understanding of galaxy formation and evolution through accurate photometric measurements. We introduce a two-step deconvolution framework using a Swin Transformer architecture. Our study reveals that the deep learning-based solution introduces a bias, constraining the scope of scientific analysis. To address this limitation, we propose a novel third step relying on the active coefficients in the sparsity wavelet framework. We conducted a performance comparison between our deep learning-based method and Firedec, a classical deconvolution algorithm, based on an analysis of a subset of the EDisCS cluster samples. We demonstrate the advantage of our method in terms of resolution recovery, generalisation to different noise properties, and computational efficiency. The analysis of this cluster sample not only allowed us to assess the efficiency of our method, but it also enabled us to quantify the number of clumps within these galaxies in relation to their disc colour. This robust technique that we propose holds promise for identifying structures in the distant universe through ground-based images.
Paper Structure (21 sections, 4 equations, 14 figures, 1 table)

This paper contains 21 sections, 4 equations, 14 figures, 1 table.

Table of Contents

  1. Introduction
  2. Deep learning-based deconvolution
  3. Tikhonov deconvolution
  4. SUNet denoising
  5. Debiasing with multi-resolution support
  6. Dataset and experiments
  7. Training dataset generation
  8. Training the SUNet
  9. Test dataset
  10. Results
  11. Objects size and clump detection
  12. Comparison with classical methods
  13. SUNet deconvolution results
  14. Resolution recovery
  15. False positives and false negatives
  16. ...and 6 more sections

Figures (14)

  • Figure 1: SUNet architecture with Swin Transformer blocks replacing the convolutional layers while preserving the multi-scale Unet backbone. Credits: sunet
  • Figure 2: Iterative recovery of lost flux through debiasing using multi-resolution support. (\ref{['subfig:MRSupp_2']}): Original SUNet output. The red square highlights the residual flux lost. (\ref{['subfig:MRSupp_1']}): Multi-resolution support matrices at each decomposed scale. (\ref{['subfig:MRSupp_3']}): Debiased solution after iterative correction with multi-resolution support highlighting the reduction in structured residuals. (\ref{['subfig:MRSupp_4']}): Standard deviation of the residual within the highlighted region as a function of the number of iterations. The process was stopped upon achieving convergence.
  • Figure 3: Distribution of the galaxy magnitudes in the HST F814W filter for the EDisCS samples, which were solely observed in the F814W filter for HST.
  • Figure 4: Size detection (outer contour) and clump detection (inner contours) using SCARLET. The first row shows the FORS2 images in the $V$-, $R$-, and $I$-bands, with the corresponding SUNet outputs displayed directly below. For comparison, the HST image in the F814W filter is shown adjacent to the SUNet $I$-band output. All images are decomposed into five scales, with the fourth scale chosen for size detection and the second scale for clump detection
  • Figure 5: SSIM between the $I$-band deconvolved outputs and the HST images in the F814W filter as a function of object magnitude. An SSIM of one implies identical images.
  • ...and 9 more figures