Deep Learning Based Rapid Tracking of Electromagnetic Radiation Source Position with Multiple Radiation Spectroscopy Detectors

Rapid and accurate tracking of radiation sources during electromagnetic radiation emergencies is essential for minimizing human exposure and enabling prompt evacuation. In this study, we propose a deep learning–based electromagnetic radiation source tracking system using multiple NaI(Tl) radiation spectroscopy detectors. The training data was constructed via GATE simulation, and the coefficients measured from three detectors were converted into ratios to compensate for various differences in conditions between simulation and experimental data. A deep neural network model was designed and trained with these ratio-based datasets, and subsequently validated with experimental data acquired using Cs-137 sources and NaI(Tl) detectors. The trained model successfully predicted the X- and Y-coordinates of radiation sources with high accuracy. The deep learning– based localization achieved an average positional accuracy of 95.65 ± 2.65% in the experimental results, with accuracies exceeding 99% at certain positions. These findings confirm that the proposed deep learning approach enables rapid and accurate electromagnetic radiation source localization, with potential applicability to real-time electromagnetic radiation emergency response.