The invention discloses a gamma-
ray energy spectrum analysis method based on the approximation coefficient and
deep learning. The gamma-
ray energy spectrum analysis method comprises following steps of modeling a gamma-
ray detector by means of a monte carlo method, and simulating an interested
nuclide energy spectrum to obtain a simulated energy spectrum; measuring the energy spectrum by the gamma-ray
detector, carrying out
smoothing processing on the energy spectrum, carrying out background rejection according to the time scale, and obtaining a pure count spectrum; extracting the approximation coefficient of the simulated energy spectrum by means of a
wavelet decomposition method, carrying out normalization
processing on the approximation coefficient of the simulated energy spectrum, extracting the approximation coefficient of the pure count spectrum by means of the
wavelet decomposition method, and carrying out normalization
processing on the approximation coefficient of the pure count spectrum; and taking the approximation coefficient of the simulated energy spectrum as a training sample of a
deep learning network in order to predict the
nuclide composition in the power spectrum actually measured by the gamma-ray
detector. According to the invention, by extracting the approximation coefficient of the simulated energy spectrum and training the
deep learning network by means of the
simulation sample, and employing the
simulation sample to predict the
nuclide composition of the actually measured energy spectrum, the energy spectrum nuclides can be identified quickly and stably.