Discovering New Strong Gravitational Lenses in the DESI Legacy Imaging Surveys

Abstract

We have conducted a search for new strong gravitational lensing systems in the Dark Energy Spectroscopic Instrument Legacy Imaging Surveys' Data Release 8. We use deep residual neural networks, building on previous work presented by Huang et al. These surveys together cover approximately one-third of the sky visible from the Northern Hemisphere, reaching a z-band AB magnitude of ∼22.5. We compile a training sample that consists of known lensing systems as well as non-lenses in the Legacy Surveys and the Dark Energy Survey. After applying our trained neural networks to the survey data, we visually inspect and rank images with probabilities above a threshold. Here we present 1210 new strong lens candidates.Unified Astronomy Thesaurus concepts: Strong gravitational lensing (1643); High-redshift galaxies (734); AGN host galaxies (2017); Galaxies (573); Galaxy clusters (584); Galaxy groups (597); Quasars (1319)

Holismokes

“…Convolutional neural networks (CNNs; LeCun et al 1998) have proven extremely efficient for pattern recognition tasks and have given a strong impetus to image analysis and processing. Recent studies largely demonstrate the ability of supervised CNNs to identify the rare gravitational lenses among large datasets (e.g., Jacobs et al 2017Jacobs et al , 2019Petrillo et al 2019;Huang et al 2021), extending previous automated algorithms (e.g., Gavazzi et al 2014;Joseph et al 2014) generally with better classification performance (Metcalf et al 2019). In Cañameras et al (2020, hereafter C20), we show that realistic simulations and careful selection of negative examples are crucial for successfully conducting a systematic search over 30 000 deg 2 with PanSTARRS multiband imaging.…”

“…In the recent past, Lanusse et al (2018) have developed ResNet architectures for lens finding on LSSTlike simulations, and obtained better results than classical CNNs on the strong lens finding challenge (Metcalf et al 2019). Subsequent studies confirm that such ResNets can efficiently select lenses on real survey data (e.g., Li et al 2020;Huang et al 2021).…”

Core Mass Estimates in Strong Lensing Galaxy Clusters Using a Single-halo Lens Model

“…In the future, new tools will expedite the current manual process of SL analysis. Examples include the introduction of convolutional neural networks for identification of SL features (e.g., Petrillo et al 2017;Jacobs et al 2019;Cañameras et al 2020;Huang et al 2021) and machine-learning algorithms to model the mass distribution of strong lenses (e.g., Bom et al 2019;Pearson et al 2019). We look forward to the continuous development of these tools, as the SHMs introduced in this work will greatly benefit from them.…”

Desenvolvendo um Ensemble de Redes Profundas para identificação de Lentes Gravitacionais: Aplicação em Regime de Poucos dados

“…Além disso conforme o volume de dados disponíveis cresce esse tipo de abordagem se torna ineficiente para análise massiva de dados. Uma alternativa a inspeção visualé o desenvolvimento de métodos computacionais de busca de sistemas de lentes automatizadas por Aprendizado de Máquina (ou Machine Learning) e Deep Learning [24,25,43,[67][68][69][70], sendo esté ultimo um ramo do Machine Learning que dispõe de redes com um grande número de camadas e parâmetros, tipicamente utilizando requerendo uso intensivo e computação e GPUs e [71]. Métodos de Deep Learning, e especificamente um subtipo, as Redes Neurais Convolucionais (da sigla em inglês CNNs) são consideradas o estado-da-arte do reconhecimento de padrões em imagens e, portanto particularmente interessantes na busca de efeito forte de lente em imagens de galáxias.…”