Pareto-Optimal Gamma Spectroscopic Radionuclide Identification Using Evolutionary Computing

Detection and identification of special nuclear materials (SNM) in measured gamma spectra is a challenge for nuclear security applications of spectroscopy. There are ongoing international efforts in industry, academia, and government to develop accurate, robust automated methods of gamma spectroscopic analysis to identify signature patterns characteristic of SNM. In this paper, we introduce a new approach for processing gamma-ray spectra measured by an NaI detector. The approach searches for radionuclide signature patterns in the measured spectrum by formulating a multiobjective optimization problem. Search in the multiobjective space is performed by an evolutionary algorithm, which identifies a solution in the context of Pareto Optimality Theory. The proposed approach is composed of two main steps: 1) signature selection, and 2) Pareto-optimal fitting. The output of the algorithm includes a list of all the identified radionuclides and their relative contributions to the gamma spectrum. The methodology is applied to a set of gamma spectra and compared to multiple linear regression fitting. Results demonstrate the superiority of the Pareto-optimal approach over multiple regression in the majority of the tested cases.