Nuclide Identification Algorithm for the Large-Size Plastic Detectors Based on Artificial Neural Network

This article proposes a multi-label classification model using an artificial neural network (ANN) to identify both an individual and a mixture of radionuclides in gamma spectra obtained from a $250\times 250\times 50$ -mm³ EJ-200 plastic scintillation detector. This model is evaluated under the scenario applied to pedestrian radiation portal monitors (RPMs) to judge how well it works in practice. The simulated and measured gamma of ²⁴¹Am, ¹³³Ba, ¹³⁷Cs, ⁶⁰Co, ¹⁵²Eu, and $^{131}\text{I}$ radioactive sources and background are used to generate the training dataset. Measurement data with varying source-to-detector distances, shielding thicknesses, and incidence angles are also taken into account. The experimental results show that the mean value of the accuracy can be achieved at about 98.8% and 94.9% for single- and multi-isotope identification, respectively. In addition, the model can well precisely recognize radionuclides in the gamma spectrum whose gain shift is up to 10%. The dependence of the true positive (TP) rate on the count quality factors of individual radionuclides, which was defined as the ratio between the net count rate and its associated uncertainty, is examined. The detection sensitivities, defined as the minimum count quality factor to obtain a TP rate of 95%, for ²⁴¹Am, ¹³³Ba, ¹³⁷Cs, ⁶⁰Co, ¹⁵²Eu, and $^{131}\text{I}$ are 8.90, 11.86, 8.96, 8.21, 12.54, and 11.89, respectively. With such encouraging results, the proposed model should be a useful technique for radionuclide recognition.