[Paper Review] SeeingGAN: Galactic image deblurring with deep learning for better morphological classification of galaxies

Fang Kai Gan, Kenji Bekki+1 more|arXiv (Cornell University)|Mar 17, 2021

Image Processing Techniques and Applications2 references4 citations

TL;DR

This paper introduces SeeingGAN, a deep learning-based generative adversarial network (GAN) that deblurs galaxy images from ground-based telescopes like Subaru to achieve Hubble Space Telescope (HST)-like clarity. By training on paired Subaru-HST images, SeeingGAN enhances fine-structure visibility (e.g., spiral arms, bars), achieving an 18% improvement in CW-SSIM over original blurred images, significantly improving morphological classification accuracy without physical telescope upgrades.

ABSTRACT

Classification of galactic morphologies is a crucial task in galactic astronomy, and identifying fine structures of galaxies (e.g., spiral arms, bars, and clumps) is an essential ingredient in such a classification task. However, seeing effects can cause images we obtain to appear blurry, making it difficult for astronomers to derive galaxies' physical properties and, in particular, distant galaxies. Here, we present a method that converts blurred images obtained by the ground-based Subaru Telescope into quasi Hubble Space Telescope (HST) images via machine learning. Using an existing deep learning method called generative adversarial networks (GANs), we can eliminate seeing effects, effectively resulting in an image similar to an image taken by the HST. Using multiple Subaru telescope image and HST telescope image pairs, we demonstrate that our model can augment fine structures present in the blurred images in aid for better and more precise galactic classification. Using our first of its kind machine learning-based deblurring technique on space images, we can obtain up to 18% improvement in terms of CW-SSIM (Complex Wavelet Structural Similarity Index) score when comparing the Subaru-HST pair versus SeeingGAN-HST pair. With this model, we can generate HST-like images from relatively less capable telescopes, making space exploration more accessible to the broader astronomy community. Furthermore, this model can be used not only in professional morphological classification studies of galaxies but in all citizen science for galaxy classifications.

Motivation & Objective

• To address the challenge of atmospheric seeing blurring ground-based galaxy images, which obscures fine structures critical for morphological classification.
• To develop a deep learning-based method that transforms blurred Subaru telescope images into HST-like deblurred images without physical telescope modifications.
• To improve the accuracy and reliability of automated and citizen science galaxy classification by enhancing image clarity.
• To demonstrate that GAN-based image translation can effectively mitigate seeing effects in astronomical imaging.
• To enable broader access to high-quality galaxy images for both professional and public astronomy projects.

Proposed method

• The method employs a conditional generative adversarial network (cGAN) trained on paired images from the Subaru telescope (blurred) and the Hubble Space Telescope (sharp).
• The generator network learns to map blurred galaxy images to deblurred, HST-like images by minimizing perceptual and adversarial losses.
• The discriminator network distinguishes between real HST images and generated deblurred images, encouraging photorealistic fidelity.
• The model is trained using a combination of pixel-wise L1 loss, perceptual loss, and adversarial loss to preserve structural and textural details.
• The training data consists of aligned Subaru-HST image pairs, with preprocessing to normalize and augment the dataset.
• The inference phase applies the trained generator to unseen Subaru images to produce deblurred outputs rapidly and efficiently.

Experimental results

Research questions

• RQ1 Can a GAN-based model effectively deblur galaxy images from ground-based telescopes to achieve HST-like clarity?
• RQ2 To what extent does deblurring improve the detection of fine galactic structures such as spiral arms, bars, and clumps?
• RQ3 How does the performance of the deblurred images compare quantitatively to original blurred images using structural similarity metrics?
• RQ4 Can this method be generalized to other ground-based telescopes like VLT or Keck with sufficient paired image data?
• RQ5 Can the deblurred images enhance the accuracy of automated and citizen science galaxy classification?

Key findings

• SeeingGAN achieves an 18% improvement in Complex Wavelet Structural Similarity Index (CW-SSIM) when comparing deblurred SeeingGAN outputs to original Subaru images, relative to HST reference images.
• The deblurred images generated by SeeingGAN show significantly enhanced visibility of fine galactic structures such as spiral arms, bars, and tidal tails, which are often indistinct in original blurred images.
• The model successfully recovers high-frequency details lost due to atmospheric seeing, demonstrating strong generalization on unseen galaxy images.
• The inference speed of SeeingGAN is fast enough to process large-scale galaxy image libraries, enabling scalable application to future surveys.
• The method is transferable: with sufficient paired data, similar models can be developed for other ground-based telescopes such as VLT, Keck, or SDSS.
• The results suggest that galaxies previously classified as S0-type due to blurred images may be reclassified as spirals upon deblurring, impacting studies of galaxy evolution and morphology density.