The challenge was to classify the 47 million astronomical objects detected by the J-PLUS project in the 3,192 degrees observed through the JAST80 telescope installed at the Observatorio Astrofísico de Javalambre (OAJ) in Teruel. This classification is necessary to be able to carry out a detailed study of the formation and evolution of galaxies, the properties of the stars of the Milky Way or the spatial distribution of quasars at cosmological distances using the twelve J-PLUS photometric bands. The tool, now used for the first time and the subject of this work, is BANNJOS, a public code developed by the team led by the astrophysicist Andrés del Pino.

BANNJOS is based on Bayesian neural networks, probabilistic models capable of finding complex patterns in multidimensional data sets such as those provided by J-PLUS. Its peculiarity is that it estimates a probability and correlations in the classification of each object as star, quasar and galaxy. BANNJOS used more than 400 photometric parameters from J-PLUS and the Gaia and WISE satellites to find the optimal way to classify the observed objects.

To get this far, BANNJOS was trained and validated using 1.5 million objects with a reliable classification, only 3% of the total J-PLUS sample, achieving a purity and completeness of more than 95%. Its application in J-PLUS has provided 20 million galaxies, one million quasars and 26 million stars, making it possible to select objects from each class with high confidence using their probabilities.

But BANNJOS is not only able to classify objects, as it is a general tool that can be trained for other investigations such as inferring atmospheric parameters on stars or distances to galaxies in J-PLUS as well as in other OAJ multi-filter surveys already underway, such as J-PAS.