Multiband image classification of astronomical objects

Abstract

Astronomy has entered the era of large digital sky surveys, transitioning from a relatively data-scarce field of study to a very data-rich one. The images coming from these new surveys are hyperspectral (having up to a few dozen bands) and noisy (due to limitations on telescope resolution and atmospheric conditions), present faint and saturated signals, and can amount to tens of terabytes. This unique set of characteristics make them very attractive for trying out deep learning methods. In this short paper, we present a multiband image classifier for stars, galaxies and quasars, and propose steps towards a semi-supervised scheme that could enable the discovery of new objects.