Files

Abstract

Low-surface-brightness galaxies (LSBGs), conventionally defined as galaxies with central surface brightness at least one magnitude less than the surface brightness of the ambient dark sky, have remain largely elusive in past wide-field surveys, since the majority of them lie below the surface brightness thresholds of those surveys. At the same time, observational and theoretical arguments point towards an LSBG-dominated galaxy population, especially in the dwarf galaxy regime. Recent observations of radially extended LSBGs, called ultra diffuse galaxies (UDGs) have posed challenges in our understanding of the galaxy formation process. New, large wide-field and deep galaxies surveys are going to shed light to this elusive low-surface-brightness regime. To do this, and due to a large amount of data they are going to generate, new analysis techniques should be developed, with machine learning providing a promising path for automating and expediting the discovery of LSBGs. In this work, I start by presenting the discovery and description of a large catalog of LSBGs (21,370 galaxies) from the first three years of the Dark Energy Survey, the largest catalog of LSBGs from a wide-field survey to date. Then I use this catalog to train a deep neural network able to accurately distinguish LSBGs from artifacts in a list of LSBG candidates. Afterward, I show how a computer vision model can be used to detect and remove spurious light reflections in astronomical images, another potential source of noise in LSBG searches. Finally, I show how neural networks can be used to automatically infer structural parameters (radius, surface brightness, etc) of galaxies, with simultaneous uncertainty quantification, faster and with similar accuracy as traditional light-profile fitting methods. I conclude with an overview of this work, discussing potential future directions and uses of the results presented in this thesis.

Details

Actions

PDF

from
to
Export
Download Full History