A New Deep Learning Model to Detect Gamma-Ray Bursts in the AGILE Anticoincidence System

Abstract

The AGILE space mission was launched in 2007 to study X-ray and gamma-ray astrophysics. AGILE operated in spinning mode from 2009 until 2024 February 14, when it re-entered the Earth’s atmosphere. This work uses data acquired from the AGILE anticoincidence system (ACS) from 2019 January 1 to 2022 December 31. The ACS is designed to reject charged background particles. It also detects X-ray photons in the 50–200 KeV energy range and saves each panel count rate in the telemetry as ratemeter data, a time series with a resolution of 1.024 s. We developed a method that uses a deep learning model to predict the background count rates of the AGILE ACS top panel (perpendicular to the pointing direction of the payload detectors) using the satellite’s orbital parameters as input. Then, we use the difference between predicted and acquired count rates to detect gamma-ray bursts (GRB). We trained the model with a background-only data set. After the training, the model can predict the ACS count rates with a mean reconstruction error of 3.8%. We used the GRBs listed in the GRBweb catalog to search for significant anomalies in the ACS data. We extracted light curves of 140 bins of 1.024 s for each GRB from the AGILE ACS to cover the trigger time of the GRBs. The model detected 39 GRBs with a significance of σ ≥ 3. The results contain four GRBs detected for the first time in the AGILE data.

Keywords

• Gamma-ray bursts
• Convolutional neural networks