LIGO Core-collapse Supernova Detection Using Convolutional Neural Networks

Abstract

Core-collapse supernovae (CCSNe) remain a critical focus in the search for gravitational waves in modern astronomy. Their detection and subsequent analysis will enhance our understanding of the explosion mechanisms in massive stars. This paper investigates the use of convolutional neural networks (CNN) to enhance the detection of gravitational waves originating from CCSNe. We employ two time–frequency analysis techniques to generate spectrograms (training data): short-time Fourier transform (STFT) and Q-transform (QT). Two CNNs were trained independently on sets of spectrogram images of simulated CCSNe signals and advanced LIGO noise. The CNNs detect CCSNe signals based on their time–frequency representation. Both CNNs achieve a near 100% true positive rate for CCSNe GW events with a signal-to-noise ratio greater than 0.5 in our test set. Nevertheless, the CNN trained on the STFT spectrograms outperforms the one based on the Q-transform for SNRs below 0.5.