Galaxies have an ‘edge’ or truncation that marks their size and gives us clues about how they form and evolve. Defining it was, until now, a tedious process involving days of work on images taken by large telescopes. A team of researchers from the University of Valladolid led by Jesús Vega-Ferrero, an astrophysicist at CEFCA, has explored the potential of a foundational Artificial Intelligence model to automate the analysis and reduce the process to seconds.

The so-called Segment Anything Model (SAM) is available to anyone, and is used to segment images and videos from everyday life. This model, developed by Meta AI, is similar to those used for automated driving or surveillance cameras.

The model was applied to a set of images of 1,047 galaxies, which had been previously labelled and measured by a manual process. The aim was to corroborate the reliability of the model and the result is that they only registered a 1% deviation in the measurement of the sizes compared to traditional methods. The images correspond to more than a thousand galaxies observed by the Hubble Space Telescope and were treated to be similar to those that will be provided by the EUCLID space mission, one of the European Space Agency's flagship missions. This mission will collect data from 1.5 billion galaxies over the next six years.

This is the first time this foundational model has been applied in the field of astrophysics. The paper published in Astronomy and Astrophysics validates its foundational use to measure galaxy sizes in large datasets in an automated way. This process requires virtually no human supervision or specific training of the model. The next step will be to apply this model to the first images already distributed by the EUCLID space mission. The paper opens up a new way of applying AI to do science. At the same time, it allows progress to be made in this field, which studies the evolution and shape of galaxies.

Más información sobre el artículo aquí.