RADIOISOTOPES

Radioisotope identification (RIID) by gamma-ray spectral analysis has been widely used, and accurate identification of radioisotopes is an important issue in various fields. Handheld instruments are commonly used for on-site RIID but often suffer from limited performance. This paper proposes a spectral deconvolution using unsupervised neural network models for RIID with handheld instruments in field use. This approach allows optimization of the neural network for deconvolution based on a measured spectrum combined with an energy-broadening matrix, and it does not require extensive training datasets or the precise modeling of the detector and measurement conditions. The performance of the proposed approach was examined in simulated and measured spectra, assuming the measurements of several radioisotopes with CsI(Tl) spectrometers. It was demonstrated that the unsupervised neural network models can improve the peak resolution more significantly compared to conventional deconvolution algorithms and contribute to RIID performance in the low energy resolution spectra.