Nuclide Spectrum Feature Extraction And Nuclide Identification Based On Deep Convolution Neural Network

Radionuclide energy spectrum identification technology has been widely used in the fields of Safety inspection for transportation of radioactive materials,nuclear waste sorting and emergency detection of nuclear.The traditional methods are usually divided into three steps: data processing,feature extraction and classification.The recognition rate is low due to the influence of background noise and detector energy resolution.Therefore,it is of great theoretical value and practical significance to study the theory and technology of fast identification of nuclides under complex background noise.This paper focuses on the methods of energy spectrum feature extraction and nuclide recognition based on deep convolution neural network(DCNN),automatically extracted the features of two-dimensional(2-D)radionuclide energy spectrum matrix to improve identification accuracy.The main contents of this paper are as follows: 1)The feature extraction methods based on the DCNN is studied.The 2-D energy spectrum is obtained by spatial mapping of one-dimensional(1-D)energy spectrum signal,and the multilayer convolutional neural network nuclide recognition model is constructed,which adaptively and implicitly extracts the deep abstract representation of the energy spectra.The feature extraction model based on different dimension DCNN is constructed.The feature extraction,classification and recognition of 1-D and 2-D energy spectrum images are carried out,and the experimental results are compared with the traditional feature extraction methods.The results show that the proposed method based on DCNN is effective for the extraction of radionuclide energy spectrum features.2)The application of DCNN in radionuclide recognition is studied.By building the DCNN model and using the acquired nuclide feature vectors to train the model parameters,form the nuclide classification model combining on-line training and off-line recognition.In order to verify the effectiveness of the proposed method,experiments are carried out on the Geant4 simulated data and the real radioactive source data.The experimental results show that the proposed method is feasible and effective.3)Nuclide identification system based on multi-feature fusion and multi-classifier fusion based on DCNN is studied.According to the multi-group DCNN model,construct the integration system from the feature extraction and classification decision-making level,achieve nuclide integration system.Compared with other traditional classification algorithms and single convolution network recognition method,the results show that integrated recognition system has the best recognition effect in the test sample set.In this paper,the method of feature extraction for energy spectra and nuclide identification based on DCNN is studied.The experimental results show that the proposed method can effectively improve the recognition accuracy of nuclides.