Isotope Identification in Low-Resolution Gamma Spectra Using Deep Learning

The traditional method of identifying radioactive isotopes in environmental samples through gamma spectroscopy using low-resolution scintillation detectors requires analysts with specialized knowledge and extensive experience in spectral processing. With low-resolution detectors and low statistical counting rates, this method has limited accuracy. Recently, with the development of computer science, deep learning models have provided a new approach to analyzing radiation spectra. In this report, we present the use of artificial neural networks to identify one or multiple radioactive isotopes through the gamma spectra of radiation sources and environmental samples such as Co-60, Cs-137, Am-241, RGTh, RGK, and RGU measured using NaI scintillation detectors at the University of Natural Sciences, Vietnam National University, Hanoi under different measurement conditions. The obtained identification results demonstrate the reliability of Convolution Neural Networks in gamma radiation spectrum analysis. (author)