A machine learning software to estimate morphological parameters of distant galaxies

Abstract

We develop a machine learning (ML) software to estimate morphological parameters (e.g., the half-light radius re) of high redshift galaxies in the Subaru/Hyper Suprime-Cam data. To make the ML software capture simultaneously galaxy morphological features and point spread function (PSF) broadening effects, we implement a two-stream convolutional neural network (CNN) for inputs of galaxy and PSF images. Thanks to large training samples of galaxy and PSF images, the two-stream CNN estimates re more accurately than a single-stream CNN with only galaxy images. Our ML software would be a useful tool to investigate galaxy morphological properties with PSF-unstable images obtained in future large-area ground-based surveys.