Application of neural networks for the analysis of gamma-ray spectra measured with a Ge spectrometer

The analysis of gamma-ray spectra to identify lines and their intensities usually requires expert knowledge and time-consuming calculations with complex fitting functions. A neural network algorithm can be applied to a gamma-ray spectral analysis owing to its excellent pattern recognition characteristics. However, a gamma-ray spectrum typically having 4096 channels is too large as a typical input data size for a neural network. We show that by applying a suitable peak search procedure, gamma-ray data can be reduced to peak energy data, which can be easily managed as input by neural networks. The method was applied to the analysis of gamma-ray spectra composed of mixed radioisotopes and the spectra of uranium ores. Radioisotope identification was successfully achieved.