Machine learning for radio galaxy morphology analysis

Doctoral Thesis

We explored how to morphologically classify well-resolved jetted radio-loud active galactic nuclei (RLAGN) in the LOw Frequency Array (LOFAR) Two-metre Sky Survey (LoTSS) using machine learning.
We investigated what morphology in total radio intensity maps can tell us about observed radio sources without complementary wavelength information and with limited visual inspection. We used a self-organising map (SOM) to model common radio morphologies and to reveal the rarest morphologies in LoTSS.
Furthermore, we turned the radio source-component association problem into an object detection problem and trained an adapted Fast region convolutional neural network to mimic the grouping of source components into unique sources as performed by astronomers for LoTSS data.
We also reduced the visual inspection required to find RLAGN remnant candidates based on their morphology, by using SOM-based features as input for a random forest classifier.
Finally, we created a machine...Show moreWe explored how to morphologically classify well-resolved jetted radio-loud active galactic nuclei (RLAGN) in the LOw Frequency Array (LOFAR) Two-metre Sky Survey (LoTSS) using machine learning.
We investigated what morphology in total radio intensity maps can tell us about observed radio sources without complementary wavelength information and with limited visual inspection. We used a self-organising map (SOM) to model common radio morphologies and to reveal the rarest morphologies in LoTSS.
Furthermore, we turned the radio source-component association problem into an object detection problem and trained an adapted Fast region convolutional neural network to mimic the grouping of source components into unique sources as performed by astronomers for LoTSS data.
We also reduced the visual inspection required to find RLAGN remnant candidates based on their morphology, by using SOM-based features as input for a random forest classifier.
Finally, we created a machine learning pipeline to identify giant radio galaxy (GRG) candidates and created a sample that contains more than ten thousand GRG. We then quantified the intrinsic GRG proper length distribution, the comoving GRG number density, and a current-day GRG lobe volume-filling fraction in clusters and filaments of the Cosmic Web.
Show less

Galaxies: active

Galaxies: peculiar

Radio continuum: galaxies

Techniques: computer vision

Methods: data analysis

Methods: statistical

Surveys

Catalogues