Analysis And Processing Of Fission Neutron Signal Of ²⁵²Cf-source-driven Nuclear Material Based On Deep Learning

As we all know,the two atomic bombs delivered by the United States during the second world war have caused great loss of timber and casualties in Hiroshima and Nagasaki in Japan.Since then,people have known and appreciated the devastating and harmfulness of nuclear weapons.To now,nuclear non-proliferation has been mentioned in many international conferences,such as Iran nuclear issue,North Korea nuclear issue and preventing terrorist organizations from possessing nuclear weapons.This shows that the prevention and control of nuclear weapons is still one of the most serious issues in the world today.Admittedly,the prevention and control of nuclear weapons can not be separated from accurate and fast nuclear arms control verification theories,instruments and methods.Among them,²⁵²Cf source driven noise analysis method in the field of nuclear arms control verification,is a kind of active measurement method which can give the information of nuclear material concentration,material quality and reactivity in the site of the verifiable field,so it is favored by people.Considering the merit,this paper based on the National Natural Science Foundation"Basic research on compressive sampling and processing technology of the ²⁵²Cf-Source-Driven neutron signal"?approval number:11605017?.Simulated the nuclear material's nuclear signal acquisition system with the help of Monte Carlo method.And introduced the deep learning into the analysis and processing of nuclear fission neutrons.Above this,the analysis and recognition research on the characteristics of nuclear material concentration,storage shape and material based on deep learning was carried out.The main research contents include:?1?The overseas and domestic research status of nuclear arms control verification technology is analyzed,the development and evolution process of ²⁵²Cf source driven noise analysis is clarified,and the working principle of nuclear material identification system?NMIS?is understood,which lays the foundation for the simulation design of the fission neutron signal measurement system.?2?Research on ²⁵²Cf source driven nuclear material concentration identification technology based on deep learning.On the basis of the working principle of the nuclear material identification system?NMIS?,the design of the fission neutron signal measurement system is completed.The design parameters are prepared and the MCNP program is written in accordance with the parameters,the running program can simulate the detection of fission neutron signals,obtain fission neutron signals with the concentration of nuclear materials,the distance of the detector and the angle of the detector in turn,and construct a lot of signals into a fission neutron signal library.Based on this,with the aid of depth learning deep convolution neural network,the training and testing of the network are completed by the training and test samples in the library,and the comparison experiments are also designed.The results show that the recognition accuracy of the deep convolution neural network is 92.5%,showing excellent performance.?3?Deep learning based ²⁵²Cf source driven nuclear material storage shape and material identification technology.Based on the nuclear material identification system?NMIS?,a new fission neutron signal measurement system is designed in view of the changes in the shape and material of nuclear materials.Based on this parameter,the MCNP program is written to simulate the fission neutron signal and to build a signal library.The signals in the library are randomly divided into training and testing samples.With the help of the Caffe toolbox of the deep convolution neural network,the training and testing of the network is completed.The experimental results show that the accuracy of identifying the shape and material of nuclear materials is 81.48%,which basically meets the requirement of accuracy.Based on the active nuclear control verification method of ²⁵²Cf source driven noise analysis and measurement method,this paper introduced deep learning theory,simulated the characteristics of particle transport by using of MCNP5 software,and carried out classification and recognition research for nuclear material concentration,storage shape,material,and had achieved good recognition results at last.The above study not only improves the accuracy of nuclear material identification in nuclear weapons control verification,but also indicates that deep learning can be effectively applied to the research of nuclear material's fission neutron signal analysis and processing.This provides a new technical method for nuclear material identification,and provides a strong theoretical basis for the identification of practical nuclear signals in the future at the same time.