Astronomical Knowledge Discovery of Very Faint Galaxies

Astronomical catalogues use the concept of quality flags to identify poor quality detections. For those cases, there is no certainty that this detection corresponds or not to an astronomical source. Normally, with future surveys using more advanced instruments, this uncertainty may be resolved. Here we present a new approach to better exploit the knowledge that can be extracted from a survey. This approach involves the combination of appropriate artificial intelligences methodologies, i.e. Bayesian inference, Voronoi tessellation and computer vision all this integrated in an automatic processing pipeline. When this pipeline is used with the COSMOS catalogue, apparent false detections are re-discovered as real sources. This demonstrates the benefits of using artificial intelligence methodologies in the exploitation of Big Data and the consequent knowledge discovery. The paper is structured as follows: section 1 presents an introduction of the problem and its context; section 2 describes the automatic processing pipeline and its specific use for the problem indicated in section 1; finally section 3 presents relevant results obtained with real data.