A new technique based on convolutional neural networks to measure the energy of protons and electrons with a single Timepix detector

Marine Ruffenach (1) , Sébastien Bourdarie (1) , Benedikt Bergmann (2) , Stefan Gohl (2) , Julien Mekki (3) , Jean-Roch Vaillé (4, 5)

1 DPHY, ONERA, Université de Toulouse [Toulouse]
2 CTU - Czech Technical University in Prague
3 CNES - Centre National d'Études Spatiales [Toulouse]
4 IES - Institut d’Electronique et des Systèmes
5 RADIAC - Radiations et composants

The Timepix chip has been exposed to the outer space for the first time with the SATRAM (Space Application of Timepix-based Radiation Monitor) instrument on Proba-V (Project for On-Board Autonomy Vegetation), a European Space Agency’s (ESA) satellite. This study’s objective is to develop a new technique to improve the separation of protons and electrons, which are detected by the single layer Timepix detector in SATRAM. The current identification method, proposed by S. Gohl et al. [1], is based on pattern recognition and stopping power measurements. In this article, the limitations of this method are discussed. A new method based on neural network trained with Geant4 data is proposed. Its validation with SATRAM data is presented. Similarly, a neural network trained with Geant4 data is introduced. Its purpose is to deduce the particles’ incident energy using the energy deposited in the Timepix.

Mots clés: Geant4; Monte-Carlo simulation; Neural network; Radiation belt; Radiation monitor; Space environment; Timepix

Domaines: Instrumentations et Détecteurs [physics.ins-det]; Physique de l'espace [physics.space-ph]

Marine Ruffenach, Sébastien Bourdarie, Benedikt Bergmann, Stefan Gohl, Julien Mekki, et al.. A new technique based on convolutional neural networks to measure the energy of protons and electrons with a single Timepix detector. IEEE Transactions on Nuclear Science, 2021, 68 (8), pp.1746-1753. ⟨10.1109/TNS.2021.3071583⟩. ⟨hal-03238974⟩