| |
 |
| Paper: |
A Multi-Task Neural Network Model for Event Reconstruction of Large Effective Area Compton Cameras |
| Volume: |
538, ADASS XXXII |
| Page: |
340 |
| Authors: |
Satoshi Takashima; Hirokazu Odaka; Hiroki Yoneda; Yuto Ichinohe; Aya Bamba; Tsuguo Aramaki; Yoshiyuki Inoue |
| DOI: |
10.26624/XJGU7333 |
| Abstract: |
Compton cameras are promising telescopes of MeV gamma-ray, the “last
window” of multi-wavelength astronomy. We have developed a multi-task neural network model to reconstruct celestial MeV gamma-ray events detected by Compton cameras to estimate initial energies and directions of incoming gamma rays. This model
can perform event reconstruction of three or more hit events even for photons that escape from a detector, which enables the telescopes to achieve an unprecedentedly large
effective area. Then, we conducted numerical experiments based on a Monte Carlo
simulation for performance evaluation. We set a 140×140×20 cm cubic liquid argon detector as a Compton camera assuming a GRAMS telescope and generated 4π
isotropic photons with energies up to 3 MeV. Our model shows excellent performance
for estimating the initial energies and the incoming directions up to eight hit events.
We also compared the reconstruction ability of our model with those of other two algorithms: a classical χ2 model and a physics-based one. The neural network model
predicts more accurately, particularly for three or four hit events. |
|
|
|
|
|
