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

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.

💡 Research Summary

This paper presents a novel integrated framework for significantly improving the automated identification of Gamma-Ray Bursts (GRBs). The core challenge addressed is the reliable detection of these transient, complex events amidst observational noise and data scarcity. The authors’ solution is twofold, combining advanced data generation with a specialized neural network architecture.

First, to overcome the severe limitation of small and imbalanced GRB datasets, the authors introduce a physics-informed data augmentation technique. They generate a large corpus of 100,000 synthetic GRB samples by strategically reducing the photon counts in real, observed GRB light curves by random proportions and then replenishing the subtracted counts with Poisson-distributed background noise. This method realistically simulates the “iceberg effect,” where faint GRB signals are submerged in noise, thereby enriching the training set with diverse examples of low-significance and short-duration bursts while crucially preserving the authentic temporal and spectral substructures of the original events.

Second, the paper proposes a new neural network model, AFFE-CNN, which integrates an Adaptive Frequency Feature Enhancement (AFFE) module into a 1D Convolutional Neural Network (CNN) based on ResNet. Recognizing that key discriminative features of GRBs are encoded in both time and frequency domains, the AFFE module transforms feature maps into the frequency domain using FFT. It then applies an attention-based adaptive filter to amplify frequency components critical for GRB identification and suppress irrelevant ones. This allows the model to explicitly capture the spectral-temporal correlations inherent in GRB light curves, a capability lacking in standard CNNs.

The proposed framework is rigorously evaluated using chronological splits of Fermi/GBM data. The AFFE-CNN model achieves outstanding classification accuracies of 97.59% on the validation set and 97.46% on the test set, substantially outperforming baseline CNNs and variants equipped with other popular feature enhancement modules like Squeeze-and-Excitation (SE) and Convolutional Block Attention Module (CBAM). The performance gain is particularly pronounced for low-SNR GRBs, demonstrating the effectiveness of the approach for challenging, faint events.

Beyond mere classification metrics, the authors perform a deep dive into the model’s learned representations. Through dimensionality reduction techniques (t-SNE and UMAP), they visualize the high-dimensional feature space. The results show not only a clear separation between GRBs and non-GRBs but, more importantly, reveal that GRBs with similar light-curve morphologies (e.g., single-pulsed, multi-pulsed) or suspected progenitor origins (e.g., supernova-associated long GRBs, kilonova-associated short GRBs) form distinct clusters within this space. This provides compelling evidence that the model has learned physically meaningful features that reflect the intrinsic properties of GRBs, transcending simple pattern recognition.

In conclusion, this work makes a significant advance in GRB identification by synergistically addressing data scarcity through physically-grounded augmentation and enhancing model capability via adaptive frequency-domain analysis. The framework sets a new state-of-the-art for detection accuracy and offers a novel, powerful diagnostic tool for probing GRB diversity and origins. It holds strong potential for application in current and future time-domain surveys, contributing to more robust multi-messenger astrophysics and early-warning systems.