Unveiling the rarest morphologies of the LOFAR Two-metre Sky Survey radio source population with self-organised maps

Abstract

Context. The Low Frequency Array (LOFAR) Two-metre Sky Survey (LoTSS) is a low-frequency radio continuum survey of the Northern sky at an unparalleled resolution and sensitivity. Aims. In order to fully exploit this huge dataset and those produced by the Square Kilometre Array in the next decade, automated methods in machine learning and data-mining will be increasingly essential both for morphological classifications and for identifying optical counterparts to the radio sources. Methods. Using self-organising maps (SOMs), a form of unsupervised machine learning, we created a dimensionality reduction of the radio morphologies for the ∼25k extended radio continuum sources in the LoTSS first data release, which is only ∼2 percent of the final LoTSS survey. We made use of PINK, a code which extends the SOM algorithm with rotation and flipping invariance, increasing its suitability and effectiveness for training on astronomical sources. Results. After training, the SOMs can be used for a wide range of science exploitation and we present an illustration of their potential by finding an arbitrary number of morphologically rare sources in our training data (424 square degrees) and subsequently in an area of the sky (∼5300 square degrees) outside the training data. Objects found in this way span a wide range of morphological and physical categories: extended jets of radio active galactic nuclei, diffuse cluster haloes and relics, and nearby spiral galaxies. Finally, to enable accessible, interactive, and intuitive data exploration, we showcase the LOFAR-PyBDSF Visualisation Tool, which allows users to explore the LoTSS dataset through the trained SOMs.

The LOFAR Two-metre Sky Survey

“…de Gasperin et al 2019, Morabito et al 2021, efficient distributed processing (Drabent et al 2019 andMechev et al 2018), photometric redshift estimators (Duncan et al 2019) and automated source classification (e.g. Mostert et al 2021 andMingo et al 2019). Excitingly, despite all of these advances, the LOFAR surveys data still retain vast, and largely untapped, potential.…”

YOUNG Star detrending for Transiting Exoplanet Recovery (YOUNGSTER) – II. Using self-organizing maps to explore young star variability in sectors 1–13 of TESS data

“…Zieba et al 2019;Tajiri et al 2020). A useful extension to this behaviour is that not only are those data which are most similar placed closest together in the final map, but those which are most different are placed furthest apart (Mostert et al 2021). However, an important proviso with the use of the SOM is that far more importance should be given to the relative position of the analysed data compared to its overall distribution, since, as noted by Geach (2012), the randomness of the initial SOM setup results in a different final distribution on each new run, even for precisely the same input data-set.…”

“…In wider astronomy SOMs have been used for a wide array of analyses, from Photometric Redshift calibration (Geach 2012;Carrasco Kind & Brunner 2014;Wright et al 2020a,b), Radio astronomy (Galvin et al 2020;Mostert et al 2021) and spectroscopic image segmentation (Fustes et al 2013;Schilliro & Romano 2021) to classification of Galaxies (Naim et al 1997;Johnston et al 2021), AGNs (Faisst et al 2019) and unusual quasars (Meusinger et al 2012). Of particular note were the discoveries of Johnston et al (2021) who showed how effective SOMs can be at identifying systematics alongside the main data-groups (in their case in a study of galaxy clustering in the Kilo-Degree Survey), and Khacef et al (2020)'s suggestion that the efficiency of SOMs could be improved for very large data-sets by applying the SOM to extracted features rather than the raw data.…”

Convolutional Deep Denoising Autoencoders for Radio Astronomical Images

“…A comprehensive review with a performance comparison of different solutions is given by Roscani et al (2020). Denoising tools are usually available within source finding software packages, like PyBDSF (Mostert et al 2021), SoFiA (Serra et al 2015) and AEGEAN (Hancock et al 2012) among the most up-to-date. Also of note are comprehensive and widely adopted software platforms, able to perform most of the previous tasks, like CASA (McMullin et al 2007), AIPS (Greisen 2003), MIRIAD (Sault et al 1995) and ASKAPsoft (Wieringa et al 2020).…”