Application Research Of LS-SVM In ? Spectrum Analysis

The ? spectrum is an intuitive result of detecting ? radiations by spectrometer.The major route getting energy and intensity of ? radiations is ? spectrum analysis.But there are some include poor self-adaptability and low accuracy in tradition spectrum analysis methods.This paper puts forward smoothing and peaks finding methods based on the Least Squares Support Vector Machines(LS-SVM)Through analyzing the development of ?spectrum analysis methods,the detecting principles of ? radiation,features of ? spectrum and the basic theory of Support Vector Machines.The goal of this research is finding a spectrum analysis method which has strong self-adaptability,better generalization ability and high accuracy.The experimental contents and conclusions are as follows:(1)The datas of ? spectrum are used as training sets.According to the distribution characteristics of ? spectrum datas,the RBF kernel function is chosen to establish the LSSVM full spectrum regression model.The penalty factors and kernel parameters is optimized by the leave-one-out verification methods.Complexity of model is reduced and accuracy of model learning is ensured.The ? spectrum is smoothed by the LS-SVM fullspectrum regression fitting smoothing method.Smoothing results show that this method can effectively eliminate noise and And the peak shape is not distorted.The D(V)vaule of the 60 Co spectrum smoothing result is 0.1105.Mean postion misregistration of peaks is0.28 channel.(2)In order to solve the problem that LS-SVM full-spectrum regression fitting smoothing method can not eliminate noise in airborne ? spectrum.This article proposes a segmented LS-SVM regression fitting method based on the distribution of energy windows.This method divides the airborne ? specturm into three energy segments according to(1~34),(34~114),(114~256)channels.The three segments are subjected to regression fitting.The data gap of segmentation is smoothed by the weighted stacking method.The smoothing results show that the segmented LS-SVM regression fitting method can effectively eliminate noise of airborne ? spectrum and peaks shape fitting well.Which overcomes the shortcomings of LS-SVM full spectrum regression fitting smoothing method in airborne ? spectrum application.The D(V)vaule is 0.065.(3)Use the first derivative of spectrum to determine the channel of all possible peaks.And extract the peak characteristic datas corresponding to these channels.The feature data is transformed according to the peak judgment condition of the traditional peak finding methods.Analyzing the new featue of data.Selcting the parameters which can be used as the criterions of peak deciding.Use this parameters as training sets after normalizedtreatment and establish LS-SVM peak presence determination classification model.The test results show that the peak presence determination method based on LS-SVM classification can accurately finds the peak position.The probability of false peaks and peaks loss is1.2%and 0 respectively.The experiment show that the LS-SVM algorithm achieves better results in smoothing and peak finding.This algorithm also has strong self-adaptability and better generalization ability.