Gamma Ray Energy Spectrum Data Processing-technology Based On Monte Carlo Method And The Neural Network Algorithm

Gamma ray energy spectrum data processing mainly includes the spectral data of smooth, peak-search, peak background deduct, peak area decomposition etc. It is one of the indispensable key technologies for gamma-ray spectrometer(or measurement system). Gamma ray spectrometer widely used in Geology & Mineral Prospecting, disease diagnosis and treatment, analysis of industrial process, radiation environmental monitoring. Moreover, with the development of modern microelectronics technology and computer technology, and with the introduction of some more complex energy spectrum data processing method and the large amount of calculation algorithm, the application and implementation of gamma-ray spectrometer has become more possible, thus the gamma-ray spectrometer performance has been improved. Especially the Na I(Tl) gamma-ray spectrometer, which is scintillation counting for gamma-ray detector, obviously has outstanding features as high efficiency, convenience in operation and maintenance, low cost, easy operating, etc. The effective ray energy spectrum data processing technology can make up for the shortage of the energy resolution(for 137 Cs 0.661 MeV generally about 7.5%).This paper is based on the research of NaI(Tl) gamma-ray spectrometer, using monte carlo simulation(MCNP5) method to establish the mono energetic photon energy spectrum response of spectrometer; building the photon energy spectrum response matrix based on the single photon energy spectrum, predicting parameters of any photon energy spectrum response matrix, combining the gamma ray spectrum data and first layer of BP network spectrum prediction parameters together to establish the second layer BP neural network model, in order to predict any single photon spectroscopy of NaI(Tl) of the measured spectral gamma-ray spectrometer under any radiator environment, so as to realize instrument spectrum decomposition; developing spectrum data processing software platform based on Monte Carlo simulation and BP neural network algorithm, thus to achieve all the functions mentioned above. This research mainly has the following achievements:① The research is conducted to establish models for different γ rays by using the MCNP5 under different conditions of instrument and measurement. The energy ranges within 0.24 MeV ~ 2.62 MeV of 43 different energies of single gamma photon energy spectrum data. Then the single photon spectrum response matrix is established through calculation and statistics.② The BP neural network inputs are redefined in this paper, the new formulas of input vector transmission, which using function as a neural network input vector and replacing the simple numerical data, are first proposed. Establishing the BP neural network training based on the 43 single energetic photon spectrum response matrix parameters and the correlated outcome is coefficient R2 > 0.95. Through the BP neural network, any single photon gamma-ray energy spectrum parameters can be realized in a certain range, fitting arbitrary incident. The research combines with the implementation of response matrix, thus it archives to fit any single gamma rays. The maximum relative error is 4.57%, the average relative error is 1.82% for each box counting energy.③ This article is based on 137 groups of experimental data. The whole spectrum of the count rate is the input vector, all single photon spectrum response parameters and the incident ray intensity are the output vectors, which establish the second layer BP neural network and training and work out the correlation coefficient R2 > 0.95, thus realize arbitrary spectrum of single photon energy spectrum- intensity distribution prediction. We can calculate the whole spectrum data according to each single photon energy spectrum response parameters, strength and other matrix data. By comparing fitting of full spectrum through the synthesis of various single photon spectrum data with the measured spectrum, we can see that the maximum rate of relative error is 4.58%, the average relative error is 1.76%; maximum relative error is 5%, the average relative error of single channel maximum count rate is 3.21%; the greatest relative error of the single channel average count rate is 5%, the average relative error is 3.58%.④ In this paper, we develop the ray energy spectrum data processing and analysis software platform based on Monte Carlo method and the improved neural network algorithm. The software can automatically calculate the single photon spectrum matrix parameters under given instrument and it can operate and display the analysis of energy spectrum, explain ray energy spectrum of given ray spectrometer.