ContinUNet: Fast deep radio image segmentation in the SKA Era with U-Net

Abstract

We present a new machine learning driven (ML) source-finding tool for next generation radio surveys that performs fast source extraction on a range of source morphologies at large dynamic ranges with minimal parameter tuning and post-processing. The construction of the Square Kilometre Array (SKA) radio telescope will revolutionise the field of radio astronomy. However, accurate and automated source-finding techniques are required to reach SKA science goals. We have developed a novel source-finding method, ContinUNet, powered by an ML segmentation algorithm, U-Net, that has proven highly effective and efficient when tested on SKA precursor data sets. Our model was trained and tested on simulated radio continuum data from SKA Science Data Challenge 1 and proved comparable to the state-of-the-art source-finding methods, PyBDSF and ProFound. ContinUNet was then tested on the MIGHTEE Early Science data without retraining and was able to extract point-like and extended sources with equal ease; processing a 1.6 deg2 field in <13 s on a supercomputer and ≈2 min on a personal laptop. We were able to associate components of extended sources without manual intervention with the powerful inference capabilities learnt within the network, making ContinUNet a promising tool for enabling science in the upcoming SKA era.