Study On Neutron/Gamma Discrimination Method Based On PCA-GA-SVM

Neutron detection technology has made a splash in recent years in practical work such as explosives detection,oil exploration,radioactive contamination detection,nuclear power generation and cancer radiotherapy.However,both in the laboratory and in practical applications,neutrons and gamma are always present together during neutron detection due to phenomena such as inelastic scattering of neutrons from the surrounding environment and radiation capture of slow neutrons,and commonly used neutron detectors are sensitive to both neutrons and gamma rays,so how to screen for neutrons and gamma rays in mixed fields that is of great importance to the research and development of neutron detection.In this study,a support vector machine(SVM)model based on principal component analysis(PCA)and genetic algorithm(GA)is proposed,taking into account the shortcomings of traditional neutron-gamma screening algorithms.This model,combined with the pulse shape screening technique,can not only perform neutron-gamma screening classification but also compensate for the shortcomings of the traditional screening algorithm very well.In this study,a neutron detection and acquisition platform is built with an astragal scintillator detector and a digital waveform acquisition module to collect the hybrid rays emitted by a small neutron source ²⁵²Cf.The data are first pre-processed using the charge comparison method and the frequency domain gradient method,and neutrons and gamma rays that satisfy both screening methods are constructed as training and test samples.The SVM model is prone to overfitting in the training and screening process due to the large dimensionality of the data.This study uses PCA to reduce the dimensionality of the data matrix,which is a good solution to the overfitting and signal stacking problems in the training and testing process.The key parameters of the SVM,the penalty factor C and the kernel function parameter g,are often taken empirically by hand,making the classification accuracy poor.The PCA-GA-SVM model was trained to achieve neutron-gamma discrimination using training samples.The PCA-GA-SVM model was then used to screen the test samples and the accuracy of the model for neutron-gamma discrimination was verified by calculating the screening error rate.The experimental results show that the PCA-GA-SVM model has a screening error rate of 0.89%for neutron signals and0.15%for gamma rays,both of which are small enough to prove that the PCA-GA-SVM model is accurate enough to screen both types of rays in mixed fields.The effectiveness of the PCA-GA-SVM model for mixed-field neutron-gamma screening was then further demonstrated by comparison with the charge comparison method and the frequency domain gradient method.Subsequently,in order to verify the optimisation effectiveness of PCA and GA for SVM,PCA-GA-SVM was compared with conventional SVM and SVM optimised by PCA and GA alone in a cross-sectional comparison.The PCA-GA-SVM model has obvious advantages in screening time and screening accuracy,demonstrating the good optimisation performance of PCA and GA for SVM.The screening model based on PCA-GA-SVM has an error rate of only 0.89%for neutron pulse screening and 0.15%for gamma-ray screening,which is the highest screening accuracy among the screening of the same number of mixed signals.Therefore,the PCA-GA-SVM model is able to distinguish neutrons and gamma rays,providing a new idea for neutron-gamma screening,which has certain application reference value for neutron-gamma-screening.