Super-resolution in astronomical images: enhancing object detection

Using paired images of JAST/T80 Telescope in OAJ and of Hubble Space Telescope we aim at increasing the resolution of JAST/T80 images with the objective of enhancing object detection. Higher resolution versions of JAST/T80 images in crowded areas of the sky will significantly increase the number of stars and/or galaxies detected, enabling a better measurement of their photometry. We pre-processed both low-resolution and high-resolution images to build a dataset of paired crops, which is used to train the Real-ESRGAN neural network. We are able to get super-resolved images with a 4× scale and a FSIM=0.8903, showing an improvement in the source detection in crowded areas. However, visual inspection of bright stars and nearby galaxies reveals that there is room for learning a better model that can reproduce the PSF and the details of the extended objects.