252Cf-source-driven nuclear material concentration identification based on deep learning
-
摘要: 针对核武器/核材料识别系统中核材料浓度识别的关键技术问题，采用Monte Carlo方法，通过建立252Cf源驱动核材料裂变中子信号样本库，模拟分析了随探测器距离和角度及核材料浓度变化的裂变脉冲中子信号特点，基于深度学习之卷积神经网络，构建了一种252Cf源驱动核材料浓度识别方法，实现了对测试样本浓度的识别，且还与BP神经网络和K最近邻方法进行了对比试验研究。结果表明，卷积神经网络算法进行核材料浓度识别，得到了高达92.05%识别准确率，不仅解决了因探测器距离和角度变化时对核材料浓度识别准确率影响的难题，而且还获得了优于BP神经网络和K最近邻算法对核材料浓度识别的认识，这为252Cf源驱动核材料浓度识别提供了一种新的途径。Abstract: For the problem of concentration identification of nuclear material in nuclear weapon/material identification system, we used the Monte Carlo method, established a database of neutron signal obtained by fission of nuclear material driven by 252Cf-source under the condition of different distance and angle of detectors. Based on the convolutional neural network in deep learning area, a method for 252Cf-source-driven nuclear material concentration identification was constructed, thereby, the identification of test samples was realized. Then a contrast experiment was conducted with the BP neural network and K-nearest neighbor method. The experimental results show that using the constructed method, a high identification rate of 92.05% is got. The problem of the accuracy of the nuclear material concentration identification was affected by the change of the distance and angle of the detector is solved, and the accuracy of this method is better than that of the BP neural network and K-nearest neighbor methods. This paper provides a new idea for the 252Cf-source-driven nuclear material concentration identification.
-
图 1 252Cf源驱动的三通道裂变中子脉冲信号测量系统示意图
Figure 1. Measurement model of 252Cf source spectrum measurement system
图 3 核材料浓度识别流程原理框图
Figure 3. Flow chart of nuclear material concentration identification
图 4 训练样本计数时域分布随角度变化示意图
Figure 4. Count distribution of training samples varies with angle
图 5 训练样本计数时域分布随浓度变化示意图
Figure 5. Count distribution of training samples varies with concentration
图 6 训练样本计数时域分布随距离变化示意图
Figure 6. Count distribution of training samples varies with distance
图 9 权值调节因子取不同值的分类错误率
Figure 9. Classification error rate varies with weight adjustment factor
图 11 卷积核取不同个数的分类错误率
Figure 11. Classification error rate varies with number of kernels
表 1 不同实验得到的分类准确率一览表
Table 1. Classification accuracy of different experiments
experiment error/% convolutional neural network 7.95 BP neural network feature Ⅰ 37.95 BP neural network feature Ⅱ 12.73 K-nearest neighbor 37.5 
下载: 导出CSV
-
[1] 刘成安, 伍钧. 核军备控制核查技术概论[M]. 北京: 国防工业出版社, 2007: 26-40.Liu Cheng'an, Wu Jun. Nuclear arms control and verification technology concept. Beijing: National Defense Industry Press, 2007: 26-40 [2] Mihalczo J T, Valentine T E, Mullens J A, et al. Physical and mathematical description of nuclear weapons identification system(NWIS) signatures[R]. The US Department of Energy Report No. Y/LB-15, 1997. [3] Mattingly J K, Valentine T E, Mihalczo J T. NWIS measurements for uranium metal annular castings[R]. The US Department of Energy Report No. Y/LB-15, 1998. [4] 冯鹏, 刘思远, 米德伶. 基于Elman神经网络的252Cf源和系统随机中子脉冲信号识别方法[J]. 强激光与粒子束, 2011, 23(8): 2224-2228. http://www.hplpb.com.cn/article/id/5395Feng Peng, Liu Siyuan, Mi Deling. Identification of stochastic neutron pulse signal of 252Cf nuclear system based on Elman neural network. High Power Laser and Particle Beams, 2011, 23(8): 2224-2228 http://www.hplpb.com.cn/article/id/5395 [5] 杨帆, 魏彪, 冯鹏, 等. 互相关及高阶谱核材料富集度识别方法[J]. 强激光与粒子束, 2013, 25(4): 1026-1030. http://www.hplpb.com.cn/article/id/7415Yang Fan, Wei Bao, Feng Peng, et al. Nuclear material enrichment identification method based on cross-correlation and high order spectra. High Power Laser and Particle Beams, 2013, 25(4): 1026-1030 http://www.hplpb.com.cn/article/id/7415 [6] 李鹏程, 魏彪, 冯鹏, 等. 基于压缩感知的252Cf源驱动核材料浓度识别技术研究[J]. 强激光与粒子束, 2015, 27: 074004. doi: 10.11884/HPLPB201527.074004Li Pengcheng, Wei Biao, Feng Peng, et al. 252Cf-source-driven nuclear material concentration identification based on compressive sensing. High Power Laser and Particle Beams, 2105, 27: 074004 doi: 10.11884/HPLPB201527.074004 [7] 李玉鑑, 张婷. 深度学习导论及案例分析[M]. 北京: 机械工业出版社, 2016: 1-116.Li Yujian, Zhang Ting. Introduction to depth learning and case analysis. Beijing: China Machine Press, 2016: 1-116 [8] Song I, Kimht J, Jeon P B. Deep learning for real-time robust facial expression recognition on a smart phone[C]//Proceedings of the 2014 IEEE International Conference on Consumer Electronics. 2014: 564-567. [9] Mesnil G, Dauphin Y, Yao K, et al. Using recurrent neural networks for slot filling in spoken language understanding[J]. IEEE Trans on Audio Speech and Language Processing, 2105, 23(3): 530-539. [10] Dahl G E, Yu D, Deng L, et al. Context-dependent pre-trained deep neural networks for large-vocabulary speech recognition[J]. IEEE Trans Audio, Speech, and Language Processing, 2012, 20(1): 30-42. doi: 10.1109/TASL.2011.2134090 [11] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//Conference on Neural Information Processing Systems. 2012: 1097-1105. [12] Sun Yi, Wang Xiaogang, Tang Xiaoou. Deep learning face representation by joint identification-verification[J]. International Conference on Neural Information Processing Systems, 2014, 27: 1988-1996. [13] 刘明, 李国军, 郝华青, 等. 基于卷积神经网络的T波形态分类[J]. 自动化学报, 2016, 42(9): 1339-1346. https://www.cnki.com.cn/Article/CJFDTOTAL-MOTO201609005.htmLiu Ming, Li Guojun, Hao Huaqing, et al. T wave shape classification based on convolutional neural network. Acta Automatica Sinica, 2016, 42(9): 1339-1346 https://www.cnki.com.cn/Article/CJFDTOTAL-MOTO201609005.htm -
点击查看大图
计量
- 文章访问数: 881
- HTML全文浏览量: 203
- PDF下载量: 291
- 被引次数: 0
