Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

In-Orbit GRB Identification Using LLM-based model for the CXPD CubeSat

Cunshi Wang, Zuke Feng, Difan Yi, Yuyang Li, Lirong Xie, Huanbo Feng, Yi Liu, Qian Liu, Yang Huang, Hongbang Liu, Xinyu Qi, Yangheng Zheng, Ali Luo, Guirong Xue, Jifeng Liu

TL;DR

This work tackles real-time GRB identification in the CXPD CubeSat's wide-FOV soft X-ray environment. It employs a multimodal large language model (miniCPM-V2.6) fine-tuned with LoRA and quantized to 4 bits, trained on Geant4-simulated 2–10 keV spectra to classify GRBs and regress spectral indices. The framework achieves 100% classification accuracy and an RMSE of 0.118 for spectral-index estimation, demonstrated within a complete onboard-ready data pipeline. This approach enables real-time transient detection and spectral analysis on CubeSats, with implications for future space missions and reduced downlink requirements.

Abstract

To validate key technologies for wide field-of-view (FOV) X-ray polarization measurements, the Cosmic X-ray Polarization Detector (CXPD) CubeSat series has been developed as a prototype platform for the Low-Energy Xray Polarization Detector (LPD) onboard the POLAR-2 mission. The wide-FOV design significantly increases the complexity of the background environment, posing notable challenges for real-time gamma-ray burst (GRB) identification. In this work, we propose an in-orbit GRB identification method based on machine learning, using simulated spectral data as input. A training dataset was constructed using a Geant4-based simulator, incorporating in-orbit background and GRB events modeled within the 2-10 keV energy range. To meet the computational constraints of onboard processing, we employ a multimodal large language model (MLLM), which is fine-tuned using low-rank adaptation (LoRA) based on miniCPM-V2.6 and quantized to 4-bit precision. The model achieves perfect classification accuracy on validation data and demonstrates strong regression performance in estimating GRB spectral indices, with an RMSE of 0.118. Furthermore, we validate the feasibility of onboard deployment through a simulated satellite data processing pipeline, highlighting the potential of our approach to enable future real-time GRB detection and spectral analysis in orbit.

In-Orbit GRB Identification Using LLM-based model for the CXPD CubeSat

TL;DR

This work tackles real-time GRB identification in the CXPD CubeSat's wide-FOV soft X-ray environment. It employs a multimodal large language model (miniCPM-V2.6) fine-tuned with LoRA and quantized to 4 bits, trained on Geant4-simulated 2–10 keV spectra to classify GRBs and regress spectral indices. The framework achieves 100% classification accuracy and an RMSE of 0.118 for spectral-index estimation, demonstrated within a complete onboard-ready data pipeline. This approach enables real-time transient detection and spectral analysis on CubeSats, with implications for future space missions and reduced downlink requirements.

Abstract

To validate key technologies for wide field-of-view (FOV) X-ray polarization measurements, the Cosmic X-ray Polarization Detector (CXPD) CubeSat series has been developed as a prototype platform for the Low-Energy Xray Polarization Detector (LPD) onboard the POLAR-2 mission. The wide-FOV design significantly increases the complexity of the background environment, posing notable challenges for real-time gamma-ray burst (GRB) identification. In this work, we propose an in-orbit GRB identification method based on machine learning, using simulated spectral data as input. A training dataset was constructed using a Geant4-based simulator, incorporating in-orbit background and GRB events modeled within the 2-10 keV energy range. To meet the computational constraints of onboard processing, we employ a multimodal large language model (MLLM), which is fine-tuned using low-rank adaptation (LoRA) based on miniCPM-V2.6 and quantized to 4-bit precision. The model achieves perfect classification accuracy on validation data and demonstrates strong regression performance in estimating GRB spectral indices, with an RMSE of 0.118. Furthermore, we validate the feasibility of onboard deployment through a simulated satellite data processing pipeline, highlighting the potential of our approach to enable future real-time GRB detection and spectral analysis in orbit.

Paper Structure

This paper contains 11 sections, 8 figures.

Table of Contents

  1. Introduction
  2. Dataset Preparation
  3. In-orbit Background
  4. GRB
  5. Methodology
  6. miniCPM
  7. Training set design
  8. Result
  9. Model validation
  10. On Satellite pipeline
  11. Discussion and Conclusion

Figures (8)

  • Figure 1: Mass model of CXPD-03 cubesat
  • Figure 2: The left panel shows the source catalog constructed from MAXI observational data. The right panel presents the simulated sky survey background map, incorporating contributions from all MAXI/GSC sources, CXB, and charged particles. The map is plotted in Galactic coordinates (with longitude increasing from left to right) and consists of 12,288 pixels, each representing a distinct incident direction.
  • Figure 3: Distribution of the GRB photon index in the 2–10 keV energy range, based on statistical analysis of observations from the Swift satellite. This distribution is used for sampling the spectral indices in the GRB detection simulations.
  • Figure 4: The energy-spectrum diagram for a GRB with power-law index equal to 1.224 (left panel) and a background (right panel) in the training set.
  • Figure 5: The left panel is the training accuracy, and the right panel is the training loss.
  • ...and 3 more figures