Applying deep learning to galaxy morphology

In observational cosmology, one of the most fundamental procedures is the morphological classification of galaxies into a taxonomy system. The challenge is to build up a robust methodology to perform a reliable classification. The main objective of this work is to investigate how to substantially improve the classification of galaxies within large datasets by mimicking human classification. We combine accurate visual classifications (from Galaxy Zoo project) with deep learning methodology. The main discrimination in galaxy morphology is the separation of galaxies between early-type from late-type. We use the GoogLeNet Inception architecture, a Deep Convolutional Neural Network with 22 layers directly applied to the images. We achieve ≈ 99% overall accuracy for 2 classes problem and ≈ 82% for 3 classes (ellpitical, non-barred spiral and barred-spiral galaxies). The result of this ongoing PhD research has potential to provide morphological classification for millions of galaxies.