Utilizing artificial neural networks to convert gamma-ray spectra from NaI(Tl) detectors to HPGe detector gamma-ray spectra

High-purity germanium radiation detectors, reknowned for their exceptional gamma-ray spectroscopy capabilities, are associated with high costs and require cooling below 77°K for optimal operation. This study introduces an innovative approach to construct high-resolution gamma-ray spectra for high-purity germanium detectors using more affordable, lower-resolution thallium-activated sodium iodide detectors. A Fully Connected Neural Network was proposed to create a mapping function for this transformation. A dataset consisting of 7200 diverse spectra was generated using measurements from single radioisotopes for both thallium-activated sodium iodide and high purity germanium detectors. The model's performance was assessed on datasets with low counts and overlapping peaks. Experimental validation was performed on 26 multiple spectra, yielding errors of 3.83% and 5.82% and accuracies of 0.998 and 0.971 for the generated and measured test data, respectively. Despite being trained solely on generated data, the model effectively converted the broad thallium-activated sodium iodide spectra into sharp high-purity germanium spectra.