YOLO–CL: Galaxy cluster detection in the SDSS with deep machine learning

¹ Université Paris Cité, CNRS(/IN2P3), Astroparticule et Cosmologie, 10 rue Alice Domon et Léonie Duquet, 75013 Paris, France
e-mail: grishin@apc.in2p3.fr; mei@apc.in2p3.fr
² Jet Propulsion Laboratory, Cahill Center for Astronomy & Astrophysics, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, California, USA
³ Université PSL, Observatoire de Paris, Sorbonne Université, CNRS, LERMA, 75 Avenue Denfert-Rochereau, 75014 Paris, France
⁴ CNES, Centre National d’Études Spatiales, 18 Av. Edouard Belin, Toulouse, France
⁵ IJCLab, Université Paris-Saclay, CNRS/IN2P3, IJCLab, rue André Ampère, Campus de la Faculté des Sciences, 91405 Orsay, France

Received: 23 January 2023
Accepted: 10 May 2023

Abstract

Galaxy clusters are powerful probes for cosmological models. Next-generation, large-scale optical and infrared surveys are poised to reach unprecedented depths and, thus, they require highly complete and pure cluster catalogs, with a well-defined selection function. We have developed a new cluster detection algorithm named YOLO for CLuster detection (YOLO–CL), which is a modified version of the state-of-the-art object detection deep convolutional network named You only look once (YOLO) that has been optimized for the detection of galaxy clusters. We trained YOLO–CL on the red-sequence Matched-filter Probabilistic Percolation (redMaPPer) cluster catalog, based on Sloan Digital Sky Survey (SDSS) color images. We find that YOLO–CL detects 95–98% of the redMaPPer clusters, with a purity of 95–98%, that is calculated by applying the network to SDSS blank fields. When compared to the Meta-Catalog of X-Ray Detected Clusters of Galaxies 2021 (MCXC2021) X-ray catalog in the SDSS footprint, YOLO–CL recovers all clusters at L_X ≳ 2–3 × 10⁴⁴ erg s⁻¹, M₅₀₀ ≳ 2–3 × 10¹⁴M_⊙, R₅₀₀≳0.75–0.8 Mpc and 0.4 ≲ z ≲ 0.6. When compared to the redMaPPer detection of the same MCXC2021 clusters, we find that YOLO–CL is more complete than redMaPPer, which means that the neural network has indeed improved the cluster detection efficiency of its training sample. In fact, YOLO–CL detects ~98% of the MCXC2021 clusters with an X-ray surface brightness of I_X,500 ≳ 20 × 10⁻¹⁵ erg s⁻¹ cm⁻² arcmin⁻² at 0.2 ≲ z ≲ 0.6 and ~100% of the MCXC2021 clusters with I_X,500 ≳ 30 × 10⁻¹⁵ erg s⁻¹ cm⁻² arcmin⁻² at 0.3 ≲ z ≲ 0.6; while redMaPPer detects ~98% of the MCXC2021 clusters with I_X,500 ≳ 55 × 10⁻¹⁵ erg s⁻¹ cm⁻² arcmin⁻² at 0.2 ≲ z ≲ 0.6 and ~100% of the MCXC2021 clusters with I_X,500 ≳ 20 × 10⁻¹⁵ erg s⁻¹ cm⁻² arcmin⁻² at 0.5 ≲ z ≲ 0.6. The YOLO–CL selection function is approximately constant with redshift, with respect to the MCXC2021 cluster X-ray surface brightness. YOLO–CL exhibits a high level of performance when compared to traditional detection algorithms applied to SDSS. Deep learning networks display a strong advantage over traditional galaxy cluster detection techniques because they do not require the galaxy’s photometric and photometric redshift catalogs. This eliminates systematic uncertainties that may be introduced during source detections and photometry, as well as photometric redshift measurements. Our results show that YOLO–CL is an efficient alternative to traditional cluster detection methods. In general, this work shows that it is worth exploring the performance of deep convolution networks for future cosmological cluster surveys, such as the Rubin/Legacy Survey of Space and Time (Rubin/LSST), Euclid, and Roman Space Telescope surveys.