CSIC Editorial, 23 March 2021
Research involving researchers from Instituto de Física de Cantabria (IFCA-CSIC-UC) and Instituto de Ciencias del Espacio (ICE-CSIC) has produced the largest morphological classification catalogue of galaxies to date, which includes
27 million galaxies. This catalogue, which combines high-quality images with artificial intelligence mechanisms, has been published in the journal
Monthly Notices of the Royal Astronomical Society (MNRAS).
Morphology of galaxies is closely related to type of stars that compose them and their formation mechanisms. Mainly, this catalogue classifies galaxies into two types of morphologies: spiral galaxies, which have a rotating disc where new stars are born, and elliptical galaxies, which are the most massive galaxies in the Universe, composed of old stars that perform random motions. Although it is easy to distinguish these two types of galaxies with the naked eye, there are two major problems: on the one hand, the large number of galaxies to be classified, which necessitated automated classifications, and, on the other hand, the fact that galaxies located at greater distances appear fainter and smaller, so images collected tended to have a lot of noise. Researchers have used data from the Dark Energy Survey (DES) - which catalogued hundreds of millions of distant galaxies over six years - and a machine learning algorithm with up to 97% accuracy to learn to classify galaxies into two types of morphologies, even the faintest and most distant galaxies.
This catalogue classifies galaxies into two types: spiral galaxies, which have a rotating disc where new stars are born, and ellipticals, the most massive galaxies in the Universe
The team downgraded high-quality images of local galaxies to what they would look like if they were more distant and introduced correct labels to train a convolutional neural network. In this way, it has been possible to learn to classify even most difficult examples. The study reports that algorithm used can get the morphology of galaxies right up to 97% of the time, regardless of level of noise and the spatial resolution of images. The use of convolutional neural networks is extremely successful for analysing and classifying images of galaxies. Convolutional neural networks are a deep learning algorithm that can receive an input image and assign a label to different aspects of image and differentiate between them.
Future implications
With this automated method, it has been possible to assign a classification to 27 million galaxies and produce the largest morphological catalogue of galaxies published to date. Some of the galaxies included in catalogue are up to 8 giga-years (Ga) apart, i.e. 8 billion years. This catalogue gives a rough picture of what galaxies were like when the Universe was half its current age, allows us to study how galaxies have changed over the past 8 Ga and observe how these structural changes are related to evolutionary paths of galaxies.
The fact that machines can learn to recognise patterns in noisy and hard-to-interpret data may have direct applications in other fields, such as security (e.g. facial recognition), image recognition industry, clinical diagnostics or climate change.
