• Acta Photonica Sinica
  • Vol. 50, Issue 3, 50 (2021)
Yaolu ZHANG1, Miao YU2, Tianying CHANG1、*, Shufan LI3, Zhifeng ZHENG4, Yue YANG1, Zhongmin WANG1, and Hongliang CUI1
Author Affiliations
  • 1College of Instrumentation & Electrical Engineering, Jilin University, Changchun3002, China
  • 2School of Electronic Information Engineering, University of Electronic Science and Technology of China,Zhongshan Institute, Zhongshan, Guangdong5840, China
  • 3Institute of Marine Science and Technology, Shandong University, Jinan250061, China
  • 4Zhuhai Pegasus Optoelectronics Technology Co.,Ltd, Zhuhai, Guangdong519000, China
  • show less
    DOI: 10.3788/gzxb20215003.0306003 Cite this Article
    Yaolu ZHANG, Miao YU, Tianying CHANG, Shufan LI, Zhifeng ZHENG, Yue YANG, Zhongmin WANG, Hongliang CUI. Phase-sensitive Optical Time-domain Reflectometric System Pattern Recognition Method Based on Wavenet[J]. Acta Photonica Sinica, 2021, 50(3): 50 Copy Citation Text show less
    Structure of φ-OTDR System
    Fig. 1. Structure of φ-OTDR System
    System of φ-OTDR pattern recognition
    Fig. 2. System of φ-OTDR pattern recognition
    Waveforms of three vibration signals act separately and simultaneously at three positions
    Fig. 3. Waveforms of three vibration signals act separately and simultaneously at three positions
    Two-dimensional time-space array diagram
    Fig. 4. Two-dimensional time-space array diagram
    Structrue of LSTM, one-dimensional convolution, causal convolution
    Fig. 5. Structrue of LSTM, one-dimensional convolution, causal convolution
    Structrue of a stack of causal convolutional layers
    Fig. 6. Structrue of a stack of causal convolutional layers
    Structure of dilated convolution multilayers
    Fig. 7. Structure of dilated convolution multilayers
    Residual network's structure
    Fig. 8. Residual network's structure
    One-dimensional CNN structure's training diagrams
    Fig. 9. One-dimensional CNN structure's training diagrams
    LSTM structure's training diagrams
    Fig. 10. LSTM structure's training diagrams
    Wavenet structure's training diagrams
    Fig. 11. Wavenet structure's training diagrams
    Network modelTrue labelPrediction labelAccuracy rate/%
    Foot stepping/%Hand tapping/%Stick striking/%
    One-dimensional CNNFoot stepping93.353.473.1892.75
    Hand tapping1.0695.473.47
    Stick striking5.445.1489.42
    LSTMFoot stepping97.890.601.5199.24
    Hand tapping0.00100.000.00
    Stick striking0.000.1599.85
    WavenetFoot stepping1000.000.0099.85
    Hand tapping0.3099.700.00
    Stick striking0.000.1599.85
    Table 1. Test sets' recognition accuracy rate of three structures
    Network modelIteration speed/(ms/step)One Epoch's time/sNumebers of training epochTraining time/sTesting time/ms
    One-dimensional CNN1226012012
    LSTM27042401 68061
    Wavenet1071669630
    Table 2. Three structures' training and testing efficiency
    Yaolu ZHANG, Miao YU, Tianying CHANG, Shufan LI, Zhifeng ZHENG, Yue YANG, Zhongmin WANG, Hongliang CUI. Phase-sensitive Optical Time-domain Reflectometric System Pattern Recognition Method Based on Wavenet[J]. Acta Photonica Sinica, 2021, 50(3): 50
    Download Citation