Identification Of Gamma/Proton Energetic Particles Based On Multi-Feature Model Design

Exploring the origin of high-energy cosmic rays is an important research task in the field of geophysics,and the identification accuracy of the components of high-energy cosmic rays plays a key role in the study of the origin of high-energy cosmic rays.Previously,most researchers used the classical characteristic variable analysis method for identification,and the identification accuracy was limited.Therefore,it is important to seek a high-precision component identification method.High-energy cosmic ray is a kind of intricate particle signal with highly abstract features,and multilateral neural network can automatically extract abstract features through multi-layer nonlinear operations,which are very suitable for highenergy cosmic ray component identification.This thesis will focus on the distribution characteristics and physical properties of secondary particles of high-energy cosmic rays,as well as the characteristic manifestations of different rays in the composition of high-energy cosmic rays,combined with traditional physical methods,and at the same time using deep neural networks,convolutional neural networks and other models.Advantages,and finally seek a multi-model fusion method.In order to fully explore the hidden features of high-energy cosmic ray components to reveal the relationship between secondary particles and cosmic ray components,a more accurate method for identifying high-energy cosmic ray components is obtained.This thesis will provide more help to find the source of high-energy cosmic rays,the cause of the formation of the "knee" region and the measurement of the complete cosmic ray energy spectrum,which is of great scientific significance.This thesis starts with the traditional physical algorithm of high-energy particle identification,and studies and analyzes the advantages and defects of the current high-energy particle identification model.For the current traditional physical methods,the recognition rate is low at low energy levels.In this thesis,a deep neural network model based on multi-physical feature parameters and a convolutional neural network model based on Shower map are presented,and a feature extraction and advantage fusion algorithm based on multi-model is proposed according to the above models.In the process of building a deep neural network model based on multi-physics feature parameters,this thesis firstly uses feature engineering to screen the corresponding physical features.This method will provide more parameter information that is conducive to high-energy particle identification.Lay the foundation.The convolutional neural network model based on the Shower map uses the distribution characteristics of the Shower map to construct the model,which is different from the traditional physical feature parameters.Compared with the feature,the case is extended from a single physical feature to the detector distribution feature,and the feature information is more abundant.The multi-model feature extraction and advantage fusion algorithm fuses two types of feature extraction models,which further increases the integrity of model feature recognition.Finally,this thesis uses the Q factor as an evaluation criterion to compare the discriminative efficiency of the three models.The model effectively solves the problem of low recognition rate of low energy levels in traditional physical models,and rationally utilizes the discrimination advantages of various models to effectively improve the recognition efficiency of gamma and proton.