A Gamma Spectrum Analysis Method Based on Approximation Coefficients and Deep Learning

Abstract

The invention discloses a gamma energy spectrum analysis method based on approximation coefficients and deep learning, comprising the following steps: using a Monte Carlo method to model a gamma detector and simulate the energy spectrum of nuclides of interest to obtain the simulated energy spectrum ; use the gamma detector to measure the energy spectrum, smooth the energy spectrum and subtract the background according to the time ratio to obtain the net count spectrum; use the wavelet decomposition method to extract the approximate coefficients of the simulated energy spectrum, and normalize the approximate coefficients of the simulated energy spectrum Normalization processing, using the wavelet decomposition method to extract the approximate coefficients of the net count spectrum, and normalize the approximate coefficients of the net count spectrum; use the approximate coefficients of the simulated energy spectrum as the training samples of the deep learning network to predict the actual measurement of the gamma detector The composition of the nuclides in the energy spectrum. The present invention extracts the approximate coefficients of the simulated energy spectrum, uses the simulated samples to train deep learning and uses them to predict the nuclide composition of the measured energy spectrum, so as to achieve fast and stable energy spectrum nuclide identification.

Description

technical field [0001] The invention belongs to the field of gamma energy spectrum analysis, in particular to a gamma energy spectrum analysis method based on approximation coefficients and deep learning. Background technique [0002] The accuracy and reliability of gamma spectrum analysis depend to a large extent on the processing of overlapping peaks. As the main carrier of gamma spectrum information, the characteristic peaks of nuclides usually overlap with each other. Due to the limitation of the energy resolution of gamma detectors and the interference of natural background radiation, it becomes very difficult to extract effective information from gamma energy spectra with overlapping peaks. Although the traditional neural network can solve this problem by inputting the full spectrum, due to the limitation of the number of hidden layers, its learning ability and predictive ability are low, and manual feature extraction is required to reduce the data dimension. Content...

AI Technical Summary

Problems solved by technology

Due to the limitation of the energy resolution of the gamma detector and the interference of natural background radiation, it becomes very difficult to extract effective information from the gamma energy spectrum with overlapping peaks

Although the traditional neural network can solve this problem by inputting the full spectrum, due to the limitation of the number of hidden layers, its learning ability and predictive ability are low, and manual feature extraction is required to reduce the data dimension.

Images