YOLO-CIANNA: Galaxy detection with deep learning in radio data

II. Winning the SKA SDC2 using a generalized 3D-YOLO network

¹ LUX, Observatoire de Paris, Université PSL, Sorbonne Université, CNRS, 75014 Paris, France
² Canadian Institute for Theoretical Astrophysics, University of Toronto, 60 St. George Street, Toronto, ON M5S 3H8, Canada
³ Research School of Astronomy & Astrophysics, Australian National University, Canberra, ACT 2610, Australia
⁴ Université de Strasbourg, CNRS UMR 7550, Observatoire astronomique de Strasbourg, 67000 Strasbourg, France
⁵ DIO, Observatoire de Paris, CNRS, PSL, 75014 Paris, France
⁶ IDRIS, CNRS, 91403 Orsay, France
⁷ Collège de France, 11 Place Marcelin Berthelot, 75005 Paris, France
⁸ Department of Physics and Electronics, Centre for Radio Astronomy Techniques and Technologies (RATT), Rhodes University, Makhanda 6140, South Africa
⁹ ORN, Observatoire de Paris, CNRS, PSL, Université d’Orléans, Nançay, France

^★ Corresponding author.

Received: 15 September 2025
Accepted: 26 January 2026

Abstract

Context. As the scientific exploitation of the Square Kilometre Array (SKA) approaches, there is a need for new advanced data analysis and visualization tools capable of processing large high-dimensional datasets.

Aims. In this study, we aim to generalize the YOLO-CIANNA deep learning source detection and characterization method for 3D hyperspectral HI emission cubes.

Methods. We present the adaptations we made to the regression-based detection formalism and the construction of an end-to-end 3D convolutional neural network (CNN) backbone. We then describe a processing pipeline for applying the method to simulated 3D HI cubes from the SKA Observatory Science Data Challenge 2 (SDC2) dataset.

Results. The YOLO-CIANNA method was originally developed and used by the MINERVA team that won the official SDC2 competition. Despite the public release of the full SDC2 dataset, no published result has yet surpassed MINERVA’s top score. In this paper, we present an updated version of our method that improves our challenge score by 9.5%. The resulting catalog exhibits a high detection purity of 92.3%, best-in-class characterization accuracy, and contains 45% more confirmed sources than concurrent classical detection tools. The method is also computationally efficient, processing the full ~1TB SDC2 data cube in 30 min on a single GPU.

Conclusions. These state-of-the-art results highlight the effectiveness of 3D CNN-based detectors for processing large hyperspectral data cubes and represent a promising step toward applying YOLO-CIANNA to observational data from SKA and its precursors.