留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于长短时记忆神经网络的能谱核素识别方法

王瑶 刘志明 万亚平 欧阳纯萍

王瑶, 刘志明, 万亚平, 等. 基于长短时记忆神经网络的能谱核素识别方法[J]. 强激光与粒子束, 2020, 32: 106001. doi: 10.11884/HPLPB202032.200118
引用本文: 王瑶, 刘志明, 万亚平, 等. 基于长短时记忆神经网络的能谱核素识别方法[J]. 强激光与粒子束, 2020, 32: 106001. doi: 10.11884/HPLPB202032.200118
Wang Yao, Liu Zhiming, Wan Yaping, et al. Energy spectrum nuclide recognition method based on long short-term memory neural network[J]. High Power Laser and Particle Beams, 2020, 32: 106001. doi: 10.11884/HPLPB202032.200118
Citation: Wang Yao, Liu Zhiming, Wan Yaping, et al. Energy spectrum nuclide recognition method based on long short-term memory neural network[J]. High Power Laser and Particle Beams, 2020, 32: 106001. doi: 10.11884/HPLPB202032.200118

基于长短时记忆神经网络的能谱核素识别方法

doi: 10.11884/HPLPB202032.200118
基金项目: 中央军委科技委创新特区项目(17-163-15-XJ-002-002-04);湖南省教育厅重点项目(17A185);湖南省自然科学基金项目(2019JJ0486);2020年度创新型省份建设专项抗击新冠肺炎疫情应急专题项目(2020SK3010)
详细信息
    作者简介:

    王 瑶(1995—),男,硕士研究生,从事核技术及应用研究;hnuscwy@163.com

    通讯作者:

    刘志明(1972—),男,教授,从事核电子学与探测技术研究;nhdxlzm@foxmail.com

  • 中图分类号: O657.62

Energy spectrum nuclide recognition method based on long short-term memory neural network

  • 摘要: 针对新兴的能谱核素识别方法在混合放射性核素的噪声环境中存在识别速度慢、准确率较低等问题,提出了基于长短时记忆神经网络(LSTM)的能谱核素识别方法。实验使用溴化镧(LaBr3)晶体探测器,分别对环境中60Co、137Cs放射性源分组测量得到能谱数据集,首先使用数据平滑方法和归一化方法进行数据预处理,然后将能谱数据按时间序列分组以获得可用的输入序列数组,最后训练LSTM模型得到预测结果。通过基于BP神经网络和卷积神经网络(CNN)的两个能谱识别模型进行对比,得到在测试集中平均识别率分别为83.45%和86.21%,而LSTM能谱识别模型平均识别率为93.04%,实验结果表明,该能谱模型在核素识别效果中表现较好,可用于快速的能谱核素识别设备上。
  • 图  1  LSTM模型结构

    Figure  1.  Long short-term memory neural network (LSTM) model structure

    图  2  基于长短时记忆神经网络的能谱核素识别模型

    Figure  2.  Spectral nuclide recognition model based on long short-term memory neural network

    图  3  LSTM能谱模型的不同学习率收敛曲线

    Figure  3.  Different learning rate convergence curves of LSTM energy spectrum model

    图  4  部分测试样本能谱数据图

    Figure  4.  Energy spectrum data graphs of part of the test samples

    图  5  BP和CNN能谱模型的学习率收敛曲线图

    Figure  5.  The learning rate convergence curve of BP and CNN energy spectrum models

    图  6  各模型的训练集损失曲线和准确率

    Figure  6.  The training set loss curve and accuracy curve of each model

    表  1  LSTM最后一个时间节点的能谱预测结果

    Table  1.   Energy spectrum prediction results of the last time node of LSTM

    sampleprediction
    60Co137Cs60Co+137Csno radioactive source
    60Co 17.25354004 −13.38463688 5.48915672 −8.59954166
    137Cs −3.00819635 6.40356064 0.29465187 2.9503994
    60Co+137Cs 0.48095784 2.74640751 8.61173725 −5.41806173
    no radioactive source −0.71746081 −1.06083262 −3.47033572 8.91043949
    下载: 导出CSV

    表  2  相同测量时间的训练集和相同测量时间的测试集结果

    Table  2.   Results of the training set with the same measurement time and the test set with the same measurement time

    sample measurement time/ssize of training set datasize of test set dataaverage accuracy/%
    5 160 40 90.15
    10 160 40 92.26
    20 160 40 92.66
    下载: 导出CSV

    表  3  混合测量时间的训练集和混合测量时间的测试集结果

    Table  3.   Results of training set and mixed measurement time test set with mixed measurement time

    sample measurement time/ssize of training set datasize of test set dataaverage accuracy/%
    5 s,10 s and 20 s mixed 480 120 92.37
    360 120 89.60
    240 120 90.96
    下载: 导出CSV

    表  4  混合测量时间的训练集和连续测量时间的测试集结果

    Table  4.   Results of mixed measurement time training set and continuous measurement time test set

    sample measurement time/ssize of training set datasize of test set dataaverage accuracy/%
    5 s,10 s and 20 s mixed 480 40 92.07
    360 40 91.57
    240 40 88.05
    下载: 导出CSV

    表  5  准确率达到100%所需训练步数和训练时长

    Table  5.   Training steps and training time required to achieve 100% accuracy

    model nametraining stepstraining time /min
    BP 67 73.10
    CNN 66 72.78
    LSTM 40 35.26
    下载: 导出CSV

    表  6  各模型识别准确率

    Table  6.   Recognition accuracy of each model

    model nameaccuracy of data set 1/%accuracy of data set 2/%accuracy of data set 3/%
    BP 83.74 83.59 83.04
    CNN 86.05 85.83 86.73
    LSTM 93.56 92.13 93.44
    下载: 导出CSV
  • [1] Liang Chen, Yi Xiangwei. Nuclide identification algorithm based on K–L transform and neural networks[J]. Nuclear Inst and Methods in Physics Research A, 2009, 598(2): 450-453.
    [2] 王一鸣, 魏义祥. 基于模糊逻辑的γ能谱核素识别[J]. 清华大学学报(自然科学版), 2012, 52(12):1736-1740. (Wang Yiming, Wei Yixiang. Fuzzy logic based nuclide identification for γ ray spectra[J]. Journal of Tsinghua University(Science and Technology), 2012, 52(12): 1736-1740
    [3] 问斯莹, 王百荣, 肖刚, 等. 基于序贯贝叶斯方法的核素识别算法研究[J]. 核电子学与探测术, 2016, 36(2):179-183. (Wen Siying, Wang Bairong, Xiao Gang, et al. The study on nuclide identification algorithm based on sequential Bayesian analysis[J]. Nuclear Electronics and Detection Technology, 2016, 36(2): 179-183
    [4] 张江梅, 季海波, 冯兴华, 等. 基于稀疏表示的核素能谱特征提取及核素识别[J]. 强激光与粒子束, 2018, 30:046003. (Zhang Jiangmei, Ji Haibo, Feng Xinghua, et al. Nuclide spectrum feature extraction and nuclide identification based on sparse representation[J]. High Power Laser and Particle Beams, 2018, 30: 046003
    [5] 胡浩行, 张江梅, 王坤朋, 等. 卷积神经网络在复杂核素识别中的应用[J]. 传感器与微系统, 2019, 38(10):154-156, 160. (Hu Haohang, Zhang Jiangmei, Wang Kunpeng, et al. Application of convolutional neural networks in identification of complex nuclides[J]. Transducer and Microsystem Technologies, 2019, 38(10): 154-156, 160
    [6] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9: 1735-1780. doi: 10.1162/neco.1997.9.8.1735
    [7] Graves A, Jaitly N, Mohamed A. Hybrid speech recognition with deep bidirectional LSTM[C]//IEEE Workshop on Automatic Speech Recognition and Understanding. 2013.
    [8] Hayashi T, Watanabe S, Toda T, et al. Duration-controlled LSTM for polyphonic sound event detection[J]. IEEE ACM Transactions on Audio, Speech and Language Processing, 2017, 25(11): 2059-2070.
    [9] 任智慧, 徐浩煜, 封松林, 等. 基于LSTM网络的序列标注中文分词法[J]. 计算机应用研究, 2017, 34(5):1321-1324, 1341. (Ren Zhihui, Xu Haoyu, Feng Songlin, et al. Sequence labeling Chinese word segmentation method based on LSTM networks[J]. Application Research of Computers, 2017, 34(5): 1321-1324, 1341
    [10] Ran J. A self-attention based LSTM network for text classification[J]. Journal of Physics: Conference Series, 2019, 1207: 12008. doi: 10.1088/1742-6596/1207/1/012008
    [11] 梁军, 柴玉梅, 原慧斌, 等. 基于极性转移和LSTM递归网络的情感分析[J]. 中文信息学报, 2015, 29(5):152-159. (Liang Jun, Chai Yumei, Yuan Huibin, et al. Polarity shifting and LSTM based recursive networks for sentiment analysis[J]. Journal of Chinese Information Processing, 2015, 29(5): 152-159
    [12] 季学武, 费聪, 何祥坤, 等. 基于LSTM网络的驾驶意图识别及车辆轨迹预测[J]. 中国公路学报, 2019, 32(6):34-42. (Ji Xuewu, Fei Cong, He Xiangkun, et al. Intention recognition and trajectory prediction for vehicles using LSTM network[J]. China Journal of Highway and Transport, 2019, 32(6): 34-42
    [13] 祝强, 李少康, 徐臻. LM算法求解大残差非线性最小二乘问题研究[J]. 中国测试, 2016, 42(3):12-16. (Zhu Qiang, Li Shaokang, Xu Zhen. Study of solving nonlinear least squares under large residual based on Levenberg-Marquardt algorithm[J]. China Measurement and Test, 2016, 42(3): 12-16
    [14] 高伟伟, 王广龙, 陈建辉, 等. 多尺度变步长最小均方自适应算法在光纤陀螺数据处理中的应用[J]. 强激光与粒子束, 2014, 26:071002. (Gao Weiwei, Wang Guanglong, Cheng Jianhui, et al. Application of multiple-scale variable step least mean square adaptive algorithm to fiber optic gyroscope data processing[J]. High Power Laser and Particle Beams, 2014, 26: 071002 doi: 10.3788/HPLPB20142607.71002
    [15] Li Q, Huang Y, Song X, et al. Moving window smoothing on the ensemble of competitive adaptive reweighted sampling algorithm[J]. Spectrochimica Acta. Part A, Molecular And Biomolecular Spectroscopy, 2019, 214: 129-138. doi: 10.1016/j.saa.2019.02.023
    [16] Bolstad B M, Irizarry R A, Astrand M, et al. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias[J]. Bioinformatics, 2003, 19(2): 185-193. doi: 10.1093/bioinformatics/19.2.185
  • 加载中
图(6) / 表(6)
计量
  • 文章访问数:  1695
  • HTML全文浏览量:  287
  • PDF下载量:  84
  • 被引次数: 0
出版历程
  • 收稿日期:  2020-05-12
  • 修回日期:  2020-08-28
  • 刊出日期:  2020-09-29

目录

    /

    返回文章
    返回