Gupta Nikhel ; Hayder Zeeshan ; Norris Ray P. ; Huynh Minh ; Petersson Lars

Abstract

Creating radio galaxy catalogues from next-generation deep surveys requiresautomated identification of associated components of extended sources and theircorresponding infrared hosts. In this paper, we introduce RadioGalaxyNET, amultimodal dataset, and a suite of novel computer vision algorithms designed toautomate the detection and localization of multi-component extended radiogalaxies and their corresponding infrared hosts. The dataset comprises 4,155instances of galaxies in 2,800 images with both radio and infrared channels.Each instance provides information about the extended radio galaxy class, itscorresponding bounding box encompassing all components, the pixel-levelsegmentation mask, and the keypoint position of its corresponding infrared hostgalaxy. RadioGalaxyNET is the first dataset to include images from the highlysensitive Australian Square Kilometre Array Pathfinder (ASKAP) radio telescope,corresponding infrared images, and instance-level annotations for galaxydetection. We benchmark several object detection algorithms on the dataset andpropose a novel multimodal approach to simultaneously detect radio galaxies andthe positions of infrared hosts.