• Acta Optica Sinica
  • Vol. 41, Issue 13, 1306019 (2021)
Zichun Zhou1、2、3, Kun Liu1、2、3、*, Junfeng Jing1、2、3, Tianhua Xu1、2、3, Shuang Wang1、2、3, Zhenshi Sun1、2、3, Hairuo Guo1、2、3, and Tiegen Liu1、2、3
Author Affiliations
  • 1School of Precision Instruments and Optoelectronics Engineering, Tianjin University, Tianjin 300072, China
  • 2Key Laboratory of Optoelectronic Information Technology of Ministry of Education, Tianjin University, Tianjin 300072, China
  • 3Institute of Optical Fiber Sensing, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/AOS202141.1306019 Cite this Article Set citation alerts
    Zichun Zhou, Kun Liu, Junfeng Jing, Tianhua Xu, Shuang Wang, Zhenshi Sun, Hairuo Guo, Tiegen Liu. Optical Fiber Vibration-Sensing Event Recognition Based on CLDNN[J]. Acta Optica Sinica, 2021, 41(13): 1306019 Copy Citation Text show less
    Schematic of DMZI distributed optical fiber vibration sensing system
    Fig. 1. Schematic of DMZI distributed optical fiber vibration sensing system
    Equivalent optical path of DMZI distributed optical fiber vibration sensor
    Fig. 2. Equivalent optical path of DMZI distributed optical fiber vibration sensor
    Network structure of CLDNN
    Fig. 3. Network structure of CLDNN
    DMZI distributed optical fiber vibration sensing system. (a) Sensing part; (b) demodulation part
    Fig. 4. DMZI distributed optical fiber vibration sensing system. (a) Sensing part; (b) demodulation part
    Preprocessing of knocking. (a) Schematic of superposition process of 5 frames of signals; (b) schematic of signal misalignment clipping after superposition
    Fig. 5. Preprocessing of knocking. (a) Schematic of superposition process of 5 frames of signals; (b) schematic of signal misalignment clipping after superposition
    Characteristic maps of 5 types of intrusion event signals. (a) No intrusion; (b) knocking; (c) crashing; (d) waggling; (e) kicking
    Fig. 6. Characteristic maps of 5 types of intrusion event signals. (a) No intrusion; (b) knocking; (c) crashing; (d) waggling; (e) kicking
    Variation of error function values of CNN in CLDNN under three convolution levels
    Fig. 7. Variation of error function values of CNN in CLDNN under three convolution levels
    LayerKernelsizeStrideOutputdimensionFunction
    Conv5×51(100,40,40,64)Relu
    Pooling2×22(100,20,20,64)Max Pooling
    Conv5×51(100,20,20,64)Relu
    Pooling2×22(100,10,10,64)Max Pooling
    Conv5×51(100,10,10,64)Relu
    Pooling2×22(100,5,5,64)Max Pooling
    Conv5×51(100,5,5,64)Relu
    Line--(100,256)Line
    LSTM128--Tanh
    FC128(100,5)-
    Output--(100,5)Softmax
    Table 1. Structural design parameters of CLDNN
    ParameterN=20N=30N=40N=50
    Accuracy 1 /%78.388.096.695.9
    Accuracy 2 /%75.087.496.294.7
    Accuracy 3 /%72.989.396.195.1
    Time /s2034300454208982
    Table 2. Average recognition rate and average training time under different clipping conditions in three tests
    AlgorithmAverage accuracy /%Number of eventsPreprocessing time /sIdentification time /s
    CLDNN96.2455×10-50.006
    RBF-EMD85.7541.140.510
    SVM93.8240.300.300
    Table 3. Comparison of recognition results of three algorithms
    AlgorithmNo intrusionKnockingCrashingWagglingKicking
    CLDNN100.099.895.287.598.7
    Table 4. [in Chinese]
    Zichun Zhou, Kun Liu, Junfeng Jing, Tianhua Xu, Shuang Wang, Zhenshi Sun, Hairuo Guo, Tiegen Liu. Optical Fiber Vibration-Sensing Event Recognition Based on CLDNN[J]. Acta Optica Sinica, 2021, 41(13): 1306019
    Download Citation