Identification of radioactive isotopes in decommissioning of nuclear facilities using ensemble learning

Accurate assessment of residual radioactive isotopes (RIs) during decommissioning of nuclear facilities is pivotal for informed dismantling procedures. This influences the selection of appropriate dismantling methods depending on isotope types and radiation levels. Datasets are made by mixing simulation data with eight gamma-emitting RIs using Monte Carlo simulation and measurement data with six gamma-emitting RIs in laboratory environment. Ensemble learning algorithms are used to distinguish the RIs. The results show that the combination of CNN and BLSTM (Type 2, 9th) has the highest accuracy, 95.3 % in only simulation data. To verify that measurement data has similar results compared with simulation data, three datasets were made by mixing simulation and measurement data (10:0, 4:6, and 2:8). The results show that using only simulation data has the highest accuracy, and the difference in performance between the three datasets is minimal. Our proposed approach is to find the optimum ensemble learning and confirm the validity of the model to demonstrate potential for accurate identification of RIs in on-site dismantling concrete by utilizing spectra from NaI(Tl) scintillator. The proposed model can serve as a useful method for establishing dismantling plans.