Poster Presentation

Generative Adversarial Networks for Astronomical Image Super-Resolution: A Case Study on Galaxy Morphology from SDSS to HSC

Author(s): Akihiro Yasuda (Ritsumeikan University), Yoshiki Toba (Ritsumeikan University), Takatoshi Shibuya (Kitami Institute of Technology)

Presenter: Akihiro Yasuda (Ritsumeikan University)

Galaxy morphology serves as a fundamental probe into the physical processes driving galaxy evolution, including galaxy mergers and interactions that often trigger AGN activity. However, reliable morphological classification requires high-resolution imaging, which is currently limited in sky coverage compared with massive low-resolution catalogs. To address this limitation, in this study, we investigated the effectiveness of astronomical image super-resolution using Generative Adversarial Networks (GANs) for galaxy morphological classification. We utilized low-resolution images from the Sloan Digital Sky Survey (SDSS) and high-resolution counterparts from the HSC-SSP (s23a_wide). By adopting pix2pix, a conditional GAN framework, we trained a model to reconstruct HSC-quality images from SDSS inputs. The quality of the generated images was quantitatively evaluated using the Structural Similarity Index Measure (SSIM). Our results demonstrate that the morphological features of spiral and merging galaxies, which were indistinguishable in the original SDSS images, were successfully recovered in the generated images. Consequently, the classification results based on the super-resolved images showed broad consistency with those derived from the actual HSC data. These findings suggest that GAN-based super-resolution is a promising approach for supporting galaxy morphological classification and artificially expanding high-resolution samples.