Classifying bent radio galaxies from a mixture of point-like/extended images with Machine Learning.

Bastien, David; Oozeer, Nadeem; Somanah, Radhakhrishna

doi:10.1088/1757-899x/198/1/012013

Cited by 3 publications

(2 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These basis functions are attractive due to having analytically-defined Fourier transforms, and so can be fitted in image space and then generated directly in visibility space. They have been used in a number of novel astronomical applications including: modelling three-dimensional distribution of dust (Schechtman-Rook et al, 2012); weak lensing measurements in simulations of radio images (Bacon et al, 2014); gravitationally lensed images (Tagore & Jackson, 2016); classifying bent radio galaxies (Bastien et al, 2017). Furthermore, they can be scaled to be extended on the sky, and lend themselves to compression/truncation.…”

Section: Introductionmentioning

confidence: 99%

Modelling and peeling extended sources with shapelets: a Fornax A case study

Line,

Mitchell,

Pindor

et al. 2020

Preprint

View full text Add to dashboard Cite

To make a power spectrum (PS) detection of the 21 cm signal from the Epoch of Reionisation (EoR), one must avoid/subtract bright foreground sources. Sources such as Fornax A present a modelling challenge due to spatial structures spanning from arc seconds up to a degree. We compare modelling with multi-scale (MS) CLEAN components to 'shapelets', an alternative set of basis functions. We introduce a new image-based shapelet modelling package, SHAMFI. We also introduce a new CUDA simulation code (WODEN) to generate point source, Gaussian, and shapelet components into visibilities. We test performance by modelling a simulation of Fornax A, peeling the model from simulated visibilities, and producing a residual PS. We find the shapelet method consistently subtracts large-angular-scale emission well, even when the angular-resolution of the data is changed. We find that when increasing the angular-resolution of the data, the MS CLEAN model worsens at large angular-scales. When testing on real MWA data, the expected improvement is not seen in real data because of the other dominating systematics still present. Through further simulation we find the expected differences to be lower than obtainable through current processing pipelines. We conclude shapelets are worthwhile for subtracting extended galaxies, and may prove essential for an EoR detection in the future, once other systematics have been addressed.

show abstract

Section: Introductionmentioning

confidence: 99%

Modelling and peeling extended sources with shapelets: a Fornax A case study

Line,

Mitchell,

Pindor

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…These basis functions are attractive due to having analytically defined Fourier transforms, and so can be fitted in image space and then generated directly in visibility space. They have been used in a number of novel astronomical applications including modelling three-dimensional (3D) distribution of dust (Schechtman-Rook, Bershady, & Wood 2012), weak lensing measurements in simulations of radio images (Bacon et al 2014), gravitationally lensed images (Tagore & Jackson 2016), and classifying bent radio galaxies (Bastien, Oozeer, & Somanah 2017). Furthermore, they can be scaled to be extended on the sky, and lend themselves to compression/truncation.…”

Section: Introductionmentioning

confidence: 99%

Modelling and peeling extended sources with shapelets: A Fornax A case study

Line

Mitchell

Pindor

et al. 2020

Publ. Astron. Soc. Aust.

View full text Add to dashboard Cite

To make a power spectrum (PS) detection of the 21-cm signal from the Epoch of Reionisation (EoR), one must avoid/subtract bright foreground sources. Sources such as Fornax A present a modelling challenge due to spatial structures spanning from arc seconds up to a degree. We compare modelling with multi-scale (MS) CLEAN components to ‘shapelets’, an alternative set of basis functions. We introduce a new image-based shapelet modelling package, SHAMFI. We also introduce a new CUDA simulation code (WODEN) to generate point source, Gaussian, and shapelet components into visibilities. We test performance by modelling a simulation of Fornax A, peeling the model from simulated visibilities, and producing a residual PS. We find the shapelet method consistently subtracts large-angular-scale emission well, even when the angular resolution of the data is changed. We find that when increasing the angular resolution of the data, the MS CLEAN model worsens at large angular scales. When testing on real Murchison Widefield Array data, the expected improvement is not seen in real data because of the other dominating systematics still present. Through further simulation, we find the expected differences to be lower than obtainable through current processing pipelines. We conclude shapelets are worthwhile for subtracting extended galaxies, and may prove essential for an EoR detection in the future, once other systematics have been addressed.

show abstract

Morphological-based Classifications of Radio Galaxies Using Supervised Machine-learning Methods Associated with Image Moments

Sadeghi

Javaherian

Miraghaei

2021

View full text Add to dashboard Cite

With the advent of new high-resolution instruments for detecting and studying radio galaxies with different morphologies, the need for the use of automatic classification methods is undeniable. Here, we focused on the morphological-based classification of radio galaxies known as Fanaroff–Riley (FR) type I and type II via supervised machine-learning approaches. Galaxy images with a resolution of 5″ at 1.4 GHz provided by the Faint Images of the Radio Sky at Twenty centimeters (FIRST) survey are employed. The radial Zernike polynomials are exploited to extract image moments. Then, the rotation, translation, and scale-invariant moments of images are used to form a training set (65% of the radio galaxy sample) and a test set (the remaining 35%). The classes of the test set are determined by two classifiers: a support vector machine and a twin support vector machine (TWSVM). In addition the genetic algorithm is employed to optimize the length of moment series and to find the optimum values of the parameters of the classifiers. The labels of outputs are compared to identify the best performance classifier. To do this the confidence level of classifications is estimated by four different metrics: precision, recall, F1 score, and accuracy. All tests show that implementing TWSVM with the radial basis function as a kernel achieves a confidence level of more than 95% in grouping galaxies.

show abstract

Classifying bent radio galaxies from a mixture of point-like/extended images with Machine Learning.

Cited by 3 publications

References 6 publications

Modelling and peeling extended sources with shapelets: a Fornax A case study

Modelling and peeling extended sources with shapelets: a Fornax A case study

Modelling and peeling extended sources with shapelets: A Fornax A case study

Morphological-based Classifications of Radio Galaxies Using Supervised Machine-learning Methods Associated with Image Moments

Contact Info

Product

Resources

About