An Application Of RBF Neural Networks In Nuclear Spectrum Smoothing Filtering

As a kind of important nuclear geophysical method,measure gamma spectrum is one of the important means to address the issue of earth science.environmental science and so on.Though the detector to detect,collect gamma rays then analyze data that can be identified in the gamma ray source belongs to what elements and give the content of element or activity.Therefore,to analyze of gamma energy spectrum,mainly including the gamma energy spectrum measurement and data collection,spectrum data of spectrum analysis,etc.Intrinsic to the statistical fluctuation of detector and the influence of electronics system noise,in the nuclear spectrum data processing will produce error.In order to weaken the effects of the error,it is necessary to smooth the data.First,introduce the Radical Basis Function Neural Network,through the establishment,leaning and training,shows the RBFNN has simple structure,fast training speed and strong anti-interference ability,especially can achieve very good approximation effect.Second,RBFNN will be applied to the energy spectrum of smoothing,through the establishment of different funtion of RBFNN,the same expansion coefficient and set up the same RBFNN,the different expansion coefficient.Based on the result of smoothing,selecting the best RBFNN and expansion coefficient.Third,through the example,using the best RBFNN and the traditional least squares algorithm to smooth a gamma spectrum,using a evaluation index ‘smoothness’,making analysis and conclusion to the smooth effect.Conclusion by analyzing the results,using the RBFNN to smooth gamma spectrum has better effect than using the least squares algorithm,according to the different requirements of smooth,RBFNN can choose different parameters.Through reduce expansion coefficient to achieve a fine smooth effect;when has much data and need for a fast smooth,can choose RBFNN that be constructed by newrbe function to smooth the spectrum.