| Nowadays, both the "Iranian nuclear issue(Iran nuclear issue)" and the DPRK nuclear issue(DPRK nuclear test)" have become one of the hottest topics in international politics, which seriously affect the sensitive nerve of the potential dangers of nuclear proliferation.The potential danger of nuclear proliferation make the existence of nuclear weapons and development of nuclear technology generate a series of uncertainty factors which is a serious threat to the world peace and the survival of mankind.So, there is no doubt that nuclear arms control verification has been repeatedly involved in today’s international security relationships. Therefore, it’s very important to master an effective verification technology so as to gain equal power in nuclear arms control because this is not only relate to the sovereignty safeguard of a country, but also the vital security interests.In many nuclear arms control verification methods, the 252Cf-source-driven noise analysis for random neutron pulse is one of them and it is an active measurement method. It is especially recommended because it has been avoided the drawbacks of passive measurement method which obtain the feature information by response the natural radiation of nuclear materials or components. However, this method is designed according to the conventional Nyquist sampling theorem. The cost of nuclear arms control verification technology is very high and the identification process is non-real time, which have become a major bottleneck of its field application. For a long term, most of the researches on the field of nuclear arms control verification technology were transplanting the signal processing and recognition algorithm to the embedded devices which did not fundamentally reduce the nuclear arms control verification cost and improve the real-time of nuclear arms control verification. If we start from the changing of data acquisition and signal processing method, it may break the technical bottlenecks of nuclear arms control verification technology fundamentally.Therefore, our researches rely on the following projects such as National Natural Science Fund of China Project(NO: 61175005), project of the Central Universities fundamental research funds(NO: CDJXS10120013) and project of the National defense(NO: GFZX02040307), always try to track foreign the frontier areas. Based on nuclear arms control verification theory, instrument and technology of 252Cf-source-driven neutron signal, this thesis will break through the traditional Nyquist sampling theorem, do some first researches on nuclear arms control verification for 252Cf-source-driven neutron signal based on compressive sensing and fractal theory. On one hand, our researches try to reduce the high speed data acquisition cost, including some key technologies such as compressed sensing reconstruction algorithm, denoise algorithm and compressive signal processing etc, on the other hand, our researches also pay attention to the depth and concentration identification of nuclear materials so as to improve timeliness and accuracy of the nuclear arms control verification systems.The main research contents of this thesis including:(1) Research on the statistical characteristics simulation of 252Cf-source-driven neutron signal. As is known to all, use extremely rare metal californium(252Cf) source to drive nuclear material fission and then research on nuclear arms control verification, is a high cost and large radiation research work. So,our research is aimed to simulate the 252Cf-source-driven neutron pulse signal based on its statistical characteristics. That is, by analyzing the statistical of neutron signal, we use cyclic shift algorithm to produce high quality uniformly distributed random numbers, and then use Bernoulli model to generate the simulated neutron signal. At the same time, we analyze the simulation results of neutron signal for of different probability values under p. Our successful simulation will lays the foundation for the following chapters that research on nuclear arms control verification technology based on compressive sensing.(2) Construct a new compressive sampling and reconstruction algorithm that is suitable for the special "0, 1" sparse structure of 252Cf-source-driven neutron signal. For the special "0, 1" sparse structure characteristic of neutron signal, we use bipartite graph to represent the compressive sampling process and then add 4 constraints to generate the compressive sampling observation matrix. On this basis, we divide the compressive sampling observation value that also been called check nodes into four types and reconstruct it in stages. Then, we analyze the impact of row-weight of observation matrix and the number of compressive sampling, discuss the reconstruction algorithm complexity and reconstruction error. The compressive sampling and reconstruction algorithm of neutron signal not only make full use of the special "0, 1" sparse structure, but also get the reconstruction result is better than the l1 norm minimization method. This provides a new way or method for signal acquisition, analysis and processing of 252Cf-source-driven neutron signal.(3) Research on the compressive sensing denoise technology for 252Cf-source-driven neutron signal. For the noise characteristics of fission neutron signal, we propose a new denoise algorithm named EMDCSDN based on based on SPGL1-BPDN denoise algorithm. Our EMDCSDN algorithm makes full use of the EMD decomposition features and selects the noise imfs automatically. Therefore, our denoise algorithm is more targeted and the denoise range will be narrowed. So, the reconstruction error of compressive sensing is reduced own to these clean imfs, this will effectively improve the denoise effect of neutron signal. The results shows that the denoise effect of our EMDCSDN algorithm is better than EMD wavelet threshold algorithm, wavelet threshold algorithm and SPGL1-BPDN algorithm, it also makes the denoise signal can reflect the characteristics of nuclear material concentration and the critical dept more clearly which is more beneficial to raise the accuracy of nuclear weapons /material identification.(4) Propose an identification method of critical depth of nuclear material based on compressive sensing. For the neutron signals of different critical depth, we use fractal theory to directly extract the characteristics of compressive sampling observation value and classify these characteristics by support vector machine. For this reason, this identification method of nuclear critical depth can be called CSMFSVM. Also, we have analyzed the influence of weight factor of multifractal generalized dimension and compressive sampling ratio on the classification accuracy rate. CSMFSVM extract the fractal characteristics of compressive sampling directly without reconstruct the original signal. It can obtain the same identification results as denoise wavelet package and Hilbert marginal spectrum feature extraction method, even better than other linear transformation feature extraction method. Therefore, CSMFSVM technology can not only improve the real-time performance of the nuclear arms control verification with higher identification accuracy, but also effectively reduce the sampling cost. It provides a new way or method of nuclear critical depth online identification of the 252Cf-source-driven neutron signal.(5) Study the identification technology of nuclear material concentration based on compressive sensing. The autocorrelation function of 252Cf-source-driven neutron signal with different concentration always has the spacing characteristics which can be regarded as feature vector for identification. We combine the compressive sensing and KNN algorithm and then propose a new identification algorithm named CSKNN based on compression sampling observation value. For the noise neutron signal, CSKNN algorithm will converge at higher accurate rate when K<20 and M/N=0.2. However, if we use EMDCSDN algorithm to denoise the neutron signal samples, which makes the spacing characteristics of compressive sampling observation value become more and more obvious. Then accurate rate will be 100% when M/N=0.05 or even smaller. The result shows that the CSKNN algorithm for nuclear material concentration can effectively reduce the sampling cost, also improve the real-time identification of nuclear material concentration, which provides a theoretical basis for online nuclear material concentration identification.In all, our research is involved with neutron physics, photoelectric detection technology, computer technology, matrix analysis, statistics and probability theory, functional analysis, optimization, graph theory, fractal theory and pattern recognition, the formation of a in the multi subject crossed technique. It has formatted a new nuclear arms control verification principle and method with cross disciplines, which can effectively reduce the data acquisition cost, improve the anti-interference ability of the system, improve the real-time of nuclear arms control verification, provide a strong safeguard system for the nuclear nonproliferation treaty. This will enhance our China’s right of speak and execution in the international nuclear arms control verification and avoid relevant countries to manipulate the results of the verification. In addition, these compressive sensing technology we used will be able to expand its applications in other fields such as data acquisition and processing. |
PDF Full Text Request