Detection of morphological structures using CNNs

Project information

Project leads: José Antonio Vázquez Mata (UNAM), Héctor Hérnandez (UNAM), Gibrán Fuentes (UNAM)

Point of contact: J. Antonio Vázquez (jvazquez@astro.unam.mx)

Rubin project code: MEX-UNA-S5

Relevant working groups: Galaxy Morphology, Low-surface-brightness Science

Project status: Active

Rationale

 Galaxy morphology summarises the internal and external physical processes that lead to the present shapes of galaxies. A reliable morphological classification is therefore essential for analysing and interpreting the physical properties of galaxies.

The new generation of extragalactic surveys like LSST (Legacy Survey of Space and Time) will provide millions of catalogued galaxies in optical bands, stimulating a diversity of approaches to infer the morphological mix of galaxies. In this context, various methods are being developed and tested, among them, important citizen projects exploding human pattern recognition capabilities (e.g. GalaxyZoo), parametric or non-parametric methods, and more recently Machine Learning (ML) algorithms considering either images, or parameters estimated from the images. 

Although machine learning techniques have been applied in astronomy over the last few years, new algorithms and architectures are still developing, as this is an important task in the current and future technologies. In the sense of morphological classification of galaxies, many authors have tried to broadly classify galaxies with great results. However, classifying galaxies in the whole Hubble sequence with great accuracy is a very complex task, among other things due to the nature of the Universe, since the morphological types are not equally found in number, being the very late type galaxies (>Scd types) very difficult to observe. The imbalance in the morphological distribution makes the ML algorithms diverge or report low accuracies for these types of galaxies. This might be one of the reasons why people are currently developing unsupervised algorithms and then associate the results with the actual morphology in the Hubble sequence. In this context, we will take advantage of state-of-art computational tools, which results can be applied to other projects.

To address this problem we have focused our work on:

1. The automatic morphological classification of galaxies, and the detection of independent morphological structures like bars, rings, mergers of galaxies, etc., after applying image processing techniques  and in combination with supervised/unsupervised CNN-based models.

2.  We will be working also on the detection and classification of low surface brightness (LSB) structures, focusing mainly on tidal debris.

Results

Talks and presentations

Morphological classification of galaxies using supervised machine learning (Gabriela Aguilar-Argüello)

Talk at LSST Galaxies SC meeting, 12 June 2023

Outputs

The code developed during this project will be published and available on GitHub.

Connections to other projects

This work is very complimentary to 'Quantitative morphology with ML (CAN-CAN-S5-1)' led by Liza Sazonova.