• Optics and Precision Engineering
  • Vol. 29, Issue 9, 2189 (2021)
Man-ling TIAN1,2,*, Dong-hui LIU1,2, Xiao-min CAO1,2, and Kuang-lu YU1,2,*
Author Affiliations
  • 1Institute of Information Science, Beijing Jiaotong University, Beijing00044, China
  • 2Beijing Key Laboratory of Modern Information Science and Network Technology, Beijing100044, China
  • show less
    DOI: 10.37188/OPE.20212909.2189 Cite this Article
    Man-ling TIAN, Dong-hui LIU, Xiao-min CAO, Kuang-lu YU. Signal processing methods of phase sensitive optical time domain reflectometer:a review[J]. Optics and Precision Engineering, 2021, 29(9): 2189 Copy Citation Text show less
    Ф-OTDR system
    Fig. 1. Ф-OTDR system
    Schematic diagram of wavelet decomposition (solid line) and wavelet packet decomposition (solid line plus dashed line)
    Fig. 2. Schematic diagram of wavelet decomposition (solid line) and wavelet packet decomposition (solid line plus dashed line)
    Threshold function in wavelet denoising
    Fig. 3. Threshold function in wavelet denoising
    EMD decomposition (IMFL: the signal after the decomposition of the L layer; resL: the residual component after the decomposition of the Lth layer)
    Fig. 4. EMD decomposition (IMFL: the signal after the decomposition of the L layer; resL: the residual component after the decomposition of the Lth layer)
    Signal reconstruction process
    Fig. 5. Signal reconstruction process
    ANN network structure with single hidden layers
    Fig. 6. ANN network structure with single hidden layers
    SVM structure
    Fig. 7. SVM structure
    Combination of multiple classifiers
    Fig. 8. Combination of multiple classifiers
    CNN-based Ф-OTDR data processing scheme
    Fig. 9. CNN-based Ф-OTDR data processing scheme
    LSTM-based Ф-OTDR data processing scheme
    Fig. 10. LSTM-based Ф-OTDR data processing scheme
    特征提取类别特征提取方法
    19时域阈值判断、均值
    58归一化
    22频域STFT
    24归一化
    37STE、水平交叉率(LCR)、Welch法
    59时域+频域快速傅里叶变换(FFT)
    57归一化、差分法、WPD、奇异值分解
    26直方图、WPD、数理统计
    25WPD、ZCR、STE、经验模态分解
    28Welch法、周期图法
    38梅尔倒谱系(MFCC)、短时过零率、ST、ZCR、短时能量(STE)

    20

    56

    21

    时频域

    多尺度小波分解

    ST-FFT

    小波分解、小波包分解

    60频谱图图像特征(SIF)
    30小波能量谱分析
    23STE、LCR、功率谱分析、WD
    61经验模态分解
    29其他图像处理
    Table 1. Literature summary of feature extraction methods
    特征定义
    峰值(PK)XPK=max(X)
    最小值(Min)XMin=min(X)
    峰值差(PK-PK)XPK-PK=maxX-min(X)2
    能量(E)XE=10lgi=1nxi-X¯2
    均值(Mean)X¯=1ni=1nxi
    平均整流值(Arv)XArv=1ni=1n|xi|
    均方根(RMS)XRMS=1ni=1nxi2
    方差(var)Xvar=1ni=1nxi-x¯2
    标准差(SD)XSD=1ni=1nxi-x¯2
    Table 2. Common time domain characteristics table
    文献分类算法传感距离/km分类准确率事件种类误报率识别时间/s
    65ANN95%3
    20ANN89.19%31.75%
    21ANN6594.4%36.7%
    29RVM2097.8%3<1
    30RVM1088.6%3
    23SVM2093.8%5
    24二叉树结构的SVM25.05>94%4<4%
    19改进SVM95.72%3
    27NC-SVM25.0594.3%55.62%0.55
    56GMM4580%2
    22GMM4569.7%831.2%
    59RF98.67%2
    58RF96.58%4
    61XGBoost25.0595.9%54.1%0.093
    60ELM1093.9%51.070 6
    57F-ELM25.0595.3%54.67%<0.1
    26SVM+BPNN+贝叶斯网络6598%20%
    25

    SVM(Ⅰ级分类器)

    RF/GBDT(Ⅱ级分类器)

    10

    99.41%

    98.02%/97.89%

    2

    4

    0.69%
    28SVM+BPNN+AdaBoost25.0594.1%5
    312D CNN+SVM4093.3%40.6
    322D CNN98.02%
    33

    2D CNN

    (改进的Googlenet)

    196.67%5
    742D CNN+SVM194.17%8
    752D CNN2>98%5
    341D CNN3495.55%5
    35MS 1-D CNN4096.59%3
    77MS CNN84.67%60.017
    76AF-CNN96.7%4
    38ALSTM5094.3%50.91
    80CLDNN3397.2%38%
    81ConvLSTM4085.6%38.25
    823D CNN+ConvLSTM40>90%3~10%
    83MLSTM-CNN25.0595.7%54.3%1.20
    841D-CNNs+Bi LSTM4097%5
    Table 3. Classification algorithm performance comparison
    Man-ling TIAN, Dong-hui LIU, Xiao-min CAO, Kuang-lu YU. Signal processing methods of phase sensitive optical time domain reflectometer:a review[J]. Optics and Precision Engineering, 2021, 29(9): 2189
    Download Citation