Identification Of γ Spectrum Fingerprints Of Nuclear Materials Based On Neural Network

With the development of the application of nuclear energy and nuclear techniques, the probability of nuclear risk is growing. This has been drawn more and more attention in radiation protection research by the public around world. Radiometry technology plays an important role in nuclear safeguard. For instance, in a future nuclear reduction treaty that requires verification of irreversible dismantling of reduced nuclear warheads, radiation fingerprints can be used to label and identify the reduced warheads. Radiation fingerprints can also be used in safety monitoring of nuclear facilities or nuclear materials, and so on. It is important to study the techniques of identifying spectrum fingerprints.Each radioactive isotope has its own uniqueγspectrum whose characteristics vary with the kinds and the compositions of nuclear materials. Therefore we can define theγspectrum as the "fingerprint" of the nuclear material. Theγspectrum fingerprints can be collected by spectroscopy and analyzed by numerical methods. The traditional methods for analyzingγspectra quantitatively and qualitatively are generally based on the positions and the areas of the photoelectric peaks in the spectra.As the development of nuclear technology and its applications, the methods based on Monte Carlo simulation are developed and have been widely used, especially in complex detecting situations. Since the nineties of the twentieth century, with the need for the verification of nuclear weapons in nuclear disarmament, Russia, the United States and other countries research and develop the technology of template measurement, which can distinct the different types of nuclear weapons. From the forming mechanics and the elementary detecting principle ofγspectra, this paper introduced the concept ofγspectrum "fingerprint", and established the RBF (radial basis function) and BP artificial neural network for identifying the types and individuals of nuclear materials based on the thought of pattern recognition, and compared the identifying capabilities of the two neural networks. Meanwhile, the effects of the detection distance and the statistical fluctuation of the system upon the identifying results were discussed. The results show that the BP neural network has the capacity of identifying the types, and the RBF neural network has strong capabilities of identifying the types and the individuals of radiation sources, and has higher identification speed. The methods of identifying nuclear materials based on artificial neural network can be applied to the nuclear safeguard...