Preliminary Development of a Deep Learning-Based Gamma Spectroscopy System for High-Radiation Field Applications

Abstract

Quantifying radioactive materials during nuclear decommissioning is essential for establishing safe and efficient strategies. However, challenges such as signal distortion, pulse pile-up, data bottlenecks, increased dead time, and spectral distortion in high-radiation environments limit accurate quantification. This paper outlines the preliminary development of a gamma spectroscopy analysis system designed for the accurate and rapid quantification of radioactive materials in high-radiation environments such as those encountered during nuclear facility decommissioning. The system comprises three primary components: a LaBr3(Ce) scintillation detector producing short signals (~150 ns) to mitigate pile-up, the KAL500 data acquisition system optimised for handling large-scale data, and an integrated software suite incorporating pulse pile-up correction and a deep learning-based gamma spectroscopy model. The developed KAL500 demonstrated its ability to retain most data in high-radiation environments, outperforming commercial data acquisition systems prone to significant data loss. The performance of the gamma spectroscopy system was evaluated using point sources and gamma irradiators and demonstrated improved energy resolution, count restoration rates, and radionuclide identification accuracy under varying conditions. This study highlighted the potential of the developed gamma spectroscopy analysis system as a robust and efficient solution for radioactive material quantification in challenging, high-radiation environments.

Keywords: High radiation environment, Data acquisition system, LaBr3(Ce) scintillation detector, Pile-up correction, Gamma spectroscopy