Hostname: page-component-7b7567c8c-8p88z Total loading time: 0 Render date: 2026-11-19T10:23:13.438Z Has data issue: false hasContentIssue false

A short study on the representation of Gravitational Waves Data for Convolutional Neural Network

Published online by Cambridge University Press:  01 August 2025

M. Grespan*
Affiliation:
National Centre for Nuclear Research, Astrophysics division ul. Pasteura 7, 02-093 Warszawa
Rights & Permissions [Opens in a new window]

Abstract

Core share and HTML view are not available for this content. However, as you have access to this content, a full PDF is available via the 'Save PDF' action button.

Binary Neutron star mergers (BNS) are of particular interest for the possibility of multi-messenger astronomy. We investigate if BNS gravitational waves trigger detection should be preferably conducted in a representation instead of another, we inspect time strains, Fourier transform, and spectrograms. We recreate the data stream of the LIGO detector, which could contain noise only or signals hidden in the noise. We train a binary classifier able to distinguish if a merger is present. We create two distinguished datasets, one with the LIGO detector instrumental noise and one whitened with respect to the detector PSD. We obtain the highest accuracy for whitened spectrograms - 93% correct classifications of an external validation dataset. Fourier transforms show similar behavior in both colored and whitened datasets with an accuracy of ∼77%. Whitened time strains have a substantial increase in accuracy from the colored to the whitened dataset, respectively 50% and 73%.

Information

Type
Poster Paper
Copyright
© The Author(s), 2025. Published by Cambridge University Press on behalf of International Astronomical Union
Supplementary material: File

Grespan supplementary material

Grespan supplementary material
Download Grespan supplementary material(File)
File 13.8 MB