Classification of radioxenon spectra with deep learning algorithm.

Journal: Journal of environmental radioactivity
PMID:

Abstract

In this study, we propose for the first time a model of classification for Beta-Gamma coincidence radioxenon spectra using a deep learning approach through the convolution neural network (CNN) technique. We utilize the entire spectrum of actual data from a noble gas system in Charlottesville (USX75 station) between 2012 and 2019. This study shows that the deep learning categorization can be done as an important pre-screening method without directly involving critical limits and abnormal thresholds. Our results demonstrate that the proposed approach of combining nuclear engineering and deep learning is a promising tool for assisting experts in accelerating and optimizing the review process of clean background and CTBT-relevant samples with high classification average accuracies of 92% and 98%, respectively.

Authors

  • Sepideh Alsadat Azimi
    Amirkabir University of Technology, Faculty of Physics and Energy Engineering, No. 350, Hafez Ave, Valiasr Square, Tehran, Iran. Electronic address: azimi.ne@aut.ac.ir.
  • Hossein Afarideh
    Amirkabir University of Technology, Faculty of Physics and Energy Engineering, No. 350, Hafez Ave, Valiasr Square, Tehran, Iran. Electronic address: hafarideh@aut.ac.ir.
  • Jong-Seo Chai
    Sungkyunkwan University, College of Information & Communication Engineering, Suwon-si, South Korea. Electronic address: jschai@skku.edu.
  • Martin Kalinowski
    Preparatory Commission for the Comprehensive Nuclear-Test-Ban-Treaty Organization, Provisional Technical Secretariat, VIC, Vienna, Austria. Electronic address: Martin.KALINOWSKI@ctbto.org.
  • Abdelhakim Gheddou
    Preparatory Commission for the Comprehensive Nuclear-Test-Ban-Treaty Organization, Provisional Technical Secretariat, VIC, Vienna, Austria. Electronic address: Abdelhakim.GHEDDOU@ctbto.org.
  • Radek Hofman
    Preparatory Commission for the Comprehensive Nuclear-Test-Ban-Treaty Organization, Provisional Technical Secretariat, VIC, Vienna, Austria. Electronic address: Radek.HOFMAN@ctbto.org.