Multi-class Convolutional Neural Network approach with Multi-layer Signal Enhancement for Gravitational Wave detection from Core-Collapse Supernovae

Core-Collapse Supernova (CCSN) are one of the most anticipated sources in the upcoming fourth observation run (O4) of the network of gravitational wave (GW) detectors. Since GW from CCSN are yet to be detected, we have developed a new data analysis pipeline called the "Multi-layer Signal Enhancement using Convolutional Neural Network (CNN) and Coherent Wave Burst (cWB)", or "MuLaSecC" in short, aiming for detection of CCSN in O4. In this work we have trained a multi-class CNN model with 4 different simulated CCSN waveforms and the background glitches. The CCSN waveforms used in our analysis are Powell and Muller 2019 s18, O'connor and Couch 2018 mesa20, Radice 2019 s9 and Scheidegger 2010 R3E1AC_L. For all classes we obtained over 95% training accuracy. The multi-class CNN provides a broader training that is more efficient in detection of potential signals with unknown morphologies. We present the results of multi-class CNN classification and efficiency curves from injected signals, as well as reconstructed waveforms of the detected signals.