The Analysis of Beta-Decay Radioactive Nuclides with Energy Spectra and Neural Network

This study applied machine learning using a neural network to analyze the nuclides from a beta-ray energy spectrum. Nuclides emitting gamma and alpha-rays are identified from the peak energy of their energy spectra. The shape of a typical beta-ray spectrum is continuous and has no apparent features such as the peak structure. This feature makes it difficult to identify nuclides from their energy spectra in the field. To overcome the challenge, we applied machine learning to beta-ray energy spectra. Five nuclides,
⁹⁰
Sr,
¹³⁷
Cs,
²⁰⁴
Tl,
¹⁴⁷
Pm, and
⁶⁰
Co, were used for analysis to validate the method. A training dataset was created by Geant4 simulation. Beta-ray energy spectra were acquired using a Si-PIN photodiode detector to evaluate the performance of the machine learning in the nuclide analysis. The accuracy of the network training reached 90%. The evaluation confirmed that the trained neural network predicted the contribution of the five radionuclides within 26% on average.