| The new algorithm for spectrometry analysis proposed in this topic research was mainly targeted at processing the data of gamma spectrum,which was measured by a detector in nuclear emergency situation,with the aim of extracting information(e.g.isotopic compositions of the spectrum,radionuclide activity)from the spectrum exactly.Compared with the existing algorithm for spectrometry analysis,new algorithm has the advantages of more intelligence,less personnel intervention,high stability of feature extraction,etc.Based on the characteristics of gamma spectrum combined with current advanced computer technology,such as neural network,deep learning,fuzzy logic,etc.,the new algorithm was designed and developed,which meets the needs of modern gamma spectrum analysis.In the whole research process,the research results obtained are as follows:(1)A rapid radionuclide identification algorithm based on the DCT and BPNN was proposed.The performance of this algorithm is not affected by detection time,radionuclide activity,detection distance,number of radionuclides as long as the MDA of a single radionuclide is met.Moreover,it has a better identification performance for the spectrum of radionuclide masked by shielding material,and the extracted feature vector of the spectrum is same as the ‘ID' of the radionuclide,which has high stability.(2)An algorithm of spectrometry analysis based on approximation coefficients and DBN was proposed.The approximation coefficients extracted by this algorithm has the same shape with the original spectrum,but its dimension is 1/8 dimension of the original spectrum,which can accelerate the convergenc of DBN in the training phase.As long as the MDA of a single radionuclide is met,the algorithm can identify the isotopic compositions of spectrum measured under different measuring conditions,such as different detection time,different number of radionuclides,different detection distance,moving radionuclide,etc.Besides,the training samples of the DBN in this algorithm are composed of spectra simulated by MCNP,which indicating this algorithm is unlimited by the kinds of radionuclides and has strong practicability in radionuclide identification.(3)A fast radionuclide identification algorithm based on WHT and SAE was proposed.The 128 transform coefficients in the low frequency region extracted by this algorithm can reserve major characteristic of the spectrum,and the SAE can learn the high-level features of 128 transform coefficients,and then it can correctly predict the isotopic compositions of spectrum.Spectralintensity would be changed on a different level in a complex environment,such as different detection times,different numbers of radionuclides,different detection distances,moving radionuclides,etc.,and the normalization of 128 transform coefficients can eliminate the effect of it.(4)An algorithm for spectrum analysis based on R-L deconvolution and fuzzy logic under low background conditions was proposed.This algorithm can enhance the performance of low-resolution gamma detector for radionuclide identification,and also can accurately decompose overlapped peaks.In additions,this algorithm can analyze the naturally occurring radionuclides contained in our surrounding environment using the spectrum recorded by low-resolution Na I(Tl)detector,and extract major information from 137Cs+60Co spectrum with low activity,which has the ability to extract the information of weak peak in the spectrum.Combined with artificial intelligence,several new algorithms for spectrometry analysis proposed in this topic research prove highly effective in radionuclide identification and activity calculation.So the application of these algorithms would boost the performance of the existing radiation monitoring instruments in radionuclide identification and activity calculation,and these algorithms would also make some contribution to our country in enhancing the capability of response to nuclear emergency. |
PDF Full Text Request