The Analysis Of γ Spectrum By RBF Network

The analysis ofγspectrum is the recognition of the kinds of radioactive nuclides by analyzingγspectrum. So it is used wildly in nuclear material analysis, the security of nuclear facilities to inspection, environmental radiation monitoring, ancient fossil fixed number of year measure and so on .The traditional recognition and analysis technique ofγspectrum rely mostly on full energy peaks. But the calculation is be made difficult by the effection of electron pairing, compton scattering and background radiation. And the recognition and analysis of complexγspectrum is also effected by overlapping full energy peaks. When theγspectrum was weak, the measurement error of full energy peaks may be promoted by local analysis. The traditional full spectrum analysis method has a strict requirements ofγspectrum, and the error accumulation often is caused.The artificial neural network, is a technique of biological neural network simulation. It is based on connection of artificial neuron. It is an intelligent system which can give suitable analysis of the input signal processing. The artificial neural network has the characteristics of distributed parallel processing, nonlinearity, robustness and fault-tolerance . So it is widely used in fuzzy recognition, intelligent control and nonlinear signal processing.There are two kind of artificial neural network, BP network and RBF network. Both of them are nonlinear general function approximation device. The RBF network's partial approximation has a high convergence speed and small network scale, compared with BP network's global approximation.In this thesis, we compete the the analysis ofγspectrum with RBF network in Matlab environment, and explore the practical software editor with Visual Basic. The RBF network's training data matrix is constructed by theγspectrum of 60Co,137Cs,152Eu,22Na and background spectrum. Then we construct the network with the training data matrix, and draw masuring curve and analysis measurement error. Finally, we prove the feasibility of the method by analyzing the simulationγspectrum.