Plutonium isotopic determination from gamma-ray spectra

The use of low- and medium-resolution room-temperature detectors for the nondestructive assay of nuclear materials has widespread applications to the safeguarding of nuclear materials. The challenge to using these detectors is the inherent difficulty of the spectral analysis to determine the amount of specific nuclear materials in the measured samples. This is especially true for extracting plutonium isotopic content from low- and medium-resolution spectral lines that are not well resolved. In this paper, neural networks trained by stochastic and singular value decomposition algorithms are applied to retrieve the plutonium isotopic content from a simulated NaI spectra. The simulated sample consists of isotopes {sup 238}Pu, {sup 239}Pu, {sup 240}Pu, {sup 241}Pu, {sup 242}Pu, and {sup 241}Am. It is demonstrated that the neutral network optimized by singular value decomposition (SVD) and stochastic training algorithms is capable of estimating plutonium content consistently resulting in an average error much smaller than the error previously reported