Advancing Identification method of Gamma-Ray Bursts with Data and Feature Enhancement

TL;DR

A one-dimensional convolutional neural network integrated with an Adaptive Frequency Feature Enhancement module and physics-informed data augmentation offers a novel diagnostic tool for identifying kilonova- and supernova-associated GRB candidates, establishing criteria to enhance multi-messenger early-warning systems.

Abstract

Gamma-ray bursts (GRBs) are challenging to identify due to their transient nature, complex temporal profiles, and limited observational datasets. We address this with a one-dimensional convolutional neural network integrated with an Adaptive Frequency Feature Enhancement module and physics-informed data augmentation. Our framework generates 100,000 synthetic GRB samples, expanding training data diversity and volume while preserving physical fidelity-especially for low-significance events. The model achieves 97.46% classification accuracy, outperforming all tested variants with conventional enhancement modules, highlighting enhanced domain-specific feature capture. Feature visualization shows model focuses on deep-seated morphological features and confirms the capability of extracting physically meaningful burst characteristics. Dimensionality reduction and clustering reveal GRBs with similar morphologies or progenitor origins cluster in the feature space, linking learned features to physical properties. This perhaps offers a novel diagnostic tool for identifying kilonova- and supernova-associated GRB candidates, establishing criteria to enhance multi-messenger early-warning systems. The framework aids current time-domain surveys, generalizes to other rare transients, and advances automated detection in large-volume observational data.

Figures (2)

• Figure 1: Data augmentation procedure for GRB light curves, using GRB 230307A as observed by Fermi/GBM detector n8 (full energy channel). The top panel presents the original light curve with a peak-SNR of 52.35 $\sigma$, where the red dashed line indicates the polynomial-fitted background level (mean count rate $\sim$90 counts/bin) and the shaded area marks the $T_{90}$ interval. The lower panels show the resulting light curves after applying count rate reductions of 30%, 60%, and 90% to each time bin, with subsequent restoration of Poisson-distributed background noise at the corresponding levels. The resulting peak SNR values are indicated for each modified light curve.
• Figure 2: Histogram of full-energy band peak-SNRs for GRB samples during the data augmentation process. The solid black line represents the SNR distribution of the primary sample, scaled by a factor of 6 for visualization. The dashed gray line denotes the fitted log-normal distribution over the range [0, 25] $\sigma$. The solid red line shows the peak-SNR distribution of 100,000 uniformly sampled instances from the primary GRB samples. The solid blue line represents the peak-SNR distribution of 100,000 randomly selected GRB samples after data augmentation. The pink shaded area highlights peak-SNR values below two $\sigma$.