Author Affiliations
1Key Laboratory of Fiber Optic Sensing and Communication, Ministry of Education, University of Electronic Science and Technology of China, Chengdu , Sichuan 611731, China2Fiber Optic Sensing Research Center, Zhijiang Laboratory, Hangzhou , Zhejiang 310000, Chinashow less
Fig. 1. Principle of the DVS/DAS based on Φ-OTDR
Fig. 2. Spatio-temporal structure of the
Φ-OTDR signal
[44] Fig. 3. De-noising and anomaly detection results based on STFT. (a) Original differential trace; (b) local energy distribution along the trace; (c) local energy distribution after the background subtraction; (d) intrusion detection and location in the energy trace; (e) intrusion detection and location in the original differential trace
[45] Fig. 4. Signal-noise separation method based on multi-scale wavelet decomposition
[44] Fig. 5. Signal-noise separation results based on multi-scale wavelet decomposition. (a) Original temporal signal; (b) combined component of
a6 and
d6; (c) combined component of
d3 and
d4; (d) combined component of
d1 and
d2[44] Fig. 6. Signal-noise separation results based on multi-scale wavelet decomposition. (a) Before the signal-noise separation; (b) after the signal-noise separation
[45] Fig. 7. Mining and recognition processing flow of sequential information based on HMM
[36] Fig. 8. State transition relationship between short-term SU features
[36] Fig. 9. Common typical event signals. (a) Background noise; (b) manual digging signal; (c) machine excavation signal; (d) traffic interference; (e) forging plant noise; (f) fabricating plant noise
[36] Fig. 10. Hidden state sequence mined by HMM. (a) Background noise; (b) manual digging signal; (c) machine excavation signal; (d) traffic interference; (e) forging plant noise; (f) fabricating plant noise
[36] Fig. 11. Training losses of different CNNs
[33] Fig. 12. Classification results of different CNN
[33] Fig. 13. Classification results of 1D-CNN combined with different models
[33] Fig. 14. Ten-fold cross classification results of 1D-CNN combined with different models
[33] Fig. 15. Spatio-temporal feature extraction process based on CNN-BiLSTM
[42] Fig. 16. Visualization results of different features. (a) Artificial features; (b) 2D-CNN features; (c) BiLSTM features; (d) CNN-BiLSTM features
[42] Fig. 17. Ten-fold cross-validation results of different models
[42] Fig. 18. Recognition time of single sample
[42] Fig. 19. Spatial energy distribution characteristics with different vertical distances. (a) 6 m; (b) 14 m
[46] Fig. 20. Vertical distances estimation method based on spatial energy distribution and integrated learning model
[46] Fig. 21. Test signal of the mechanical knocking. (a) Knocking scene; (b) time domain signal diagram
[46] Fig. 22. Spatial energy attenuation curves of the machine knocking signals. (a) Group 1; (b) group 2
[46] Fig. 23. Test signal of the mechanical excavation. (a) Excavation scene; (b) time domain signal diagram
[46] Fig. 24. Spatial energy attenuation curves of the excavation signals
[46] Fig. 25. Principles of border control and security technology
[7] Fig. 26. Laying method of optical cable and the monitoring signal before and after noise removal. (a) Laying method of the optical cable; (b) monitoring signal before denoising; (c) monitoring signal after denoising
[7] Fig. 27. Monitoring site for excavation prevention of long-distance oil pipelines. (a) Monitoring equipment; (b) gas station; (c) on-site test environment
[49] Fig. 28. Characteristic radar chart of typical event in an oil pipeline. (a) Background noise; (b) manual excavation; (c) mechanical excavation; (d) traffic disturbance; (e) factory interference
Fig. 29. Principle of the pipeline optical cable anti-theft and operation and maintenance monitoring system
Fig. 30. Interface of online monitoring and inspection. (a) Online positioning and inspection based on Baidu map; (b) statistical results of optical cable information
Fig. 31. Project site of submarine cable safety monitoring. (a) Monitored marine area; (b) monitoring center; (c) monitoring setup
Fig. 32. Monitoring site and test equipment of overhead transmission cables. (a) Monitoring center; (b) monitoring setup
[49] Fig. 33. Frequency and space distribution of cable wind dance. (a) 1:00—2:00; (b) 14:00—15:00
[50] Fig. 34. Installation wiring diagram of outdoor optical cable. (a) Sectional view; (b) top view; (c) installation process
[4] Fig. 35. Leak response signal of the DVS/DAS system. (a) Response graph of leakage when the valve is not opened; (b) response graph of leakage when the valve is opened
[4] Institutional unit | Feature extraction dimension | Recognition network or model | Attention mechanism | End-to-end network | Ref. |
---|
Beijing Jiaotong University | temporal | XGBoost | no | false | [31] [32] | temporal | F-ELM | no | false | University of Electronic Science and Technology of China | temporal | 1D-CNN | no | true | [33] | University of San Pablo Central European University | temporal | GMMs | no | false | [34] [35] | temporal contextual sequence | GMMs+HMM | no | false | University of Electronic Science and Technology of China | temporal structure and contextual sequence | HMM | no | false | [36] | Tianjin University | multiscale temporal | MS-CNN+CPL | no | true | [37] | Anhui University | multiscale temporal | MS-CNN | no | true | [38] | Transportation, Security, Energy & Automation Systems Business Sector | time-frequency | 2D-CNN | no | false | [28] | Beijing Institute of Technology | time-frequency | 2D-CNN | no | false | [29] | Zhejiang University | time-frequency | 2D-CNN+SVM | no | false | [30] | Shanghai Maritime University | time-frequency | PNN | no | false | [39] | University of Cologne | time-frequency | ALSTM | yes | false | [40] | Tianjin University | spatial-temporal | 2D-CNN | no | false | [41] | University of Electronic Science and Technology of China | spatial-temporal | 1D-CNN+BiLSTM | no | true | [42] | Sichuan University | spatial-temporal | 2D-CNN+ LSTM | no | false | [43] |
|
Table 1. DVS/DAS signal detection and recognition method combined with machine learning model
Different detection method | Energy threshold detection method | Modular maximum method of wavelet transform | STFT-based method |
---|
PD/% | 76.73 | 95.65 | 98.76 | NAR(24 h) | 287 | 161 | 2 |
|
Table 2. Actual detection results of different methods
Feature type | Feature name |
---|
Time domain | main impact strength、short time average magnitude、short time average energy | Frequency domain | frequency band variance of PSD、frequency band information entropy of PSD、mean of amplitude of PSD、procrustes mean shape of PSD、amplitude standard deviation of PSD、shape standard deviation of PSD、amplitude of skewness of PSD、shape of skewness of PSD、amplitude of kurtosis of PSD、 shape of kurtosis of PSD | Transformation domain | wavelet packet energy spectrum、information entropy of wavelet packet、MFCC | Model feature | autoregression model coefficient、linear prediction model coefficient |
|
Table 3. Local structural features of the short-term SU
Model | Average recognition rate | Type | Precision | Recall | F-value |
---|
HMM | 0.982 | 1 | 1.0000 | 1.0000 | 1.0000 | 2 | 1.0000 | 1.0000 | 1.0000 | 3 | 1.0000 | 1.0000 | 1.0000 | 4 | 0.9524 | 1.0000 | 0.9756 | 5 | 1.0000 | 0.9130 | 0.9545 | SVM | 0.919 | 1 | 1.0000 | 1.0000 | 1.0000 | 2 | 0.7500 | 1.0000 | 0.8571 | 3 | 1.0000 | 1.0000 | 1.0000 | 4 | 0.8974 | 0.8750 | 0.8861 | 5 | 1.0000 | 0.8261 | 0.9048 | RF | 0.928 | 1 | 1.0000 | 0.9524 | 0.9756 | 2 | 0.8667 | 0.8667 | 0.8667 | 3 | 0.9231 | 1.0000 | 0.9600 | 4 | 0.9000 | 0.9000 | 0.9000 | 5 | 0.9130 | 0.9130 | 0.9130 | XGB | 0.937 | 1 | 0.9524 | 1.0000 | 0.9756 | 2 | 0.8667 | 1.0000 | 0.9286 | 3 | 1.0000 | 1.0000 | 1.0000 | 4 | 0.9750 | 0.8667 | 0.9176 | 5 | 0.8696 | 0.9524 | 0.9091 | DT | 0.892 | 1 | 1.0000 | 0.9524 | 0.9756 | 2 | 0.8125 | 0.8667 | 0.8387 | 3 | 1.0000 | 1.0000 | 1.0000 | 4 | 0.8611 | 0.7750 | 0.8158 | 5 | 0.7407 | 0.8696 | 0.8000 | BN | 0.783 | 1 | 0.9524 | 1.0000 | 0.9756 | 2 | 0.6667 | 0.4545 | 0.5405 | 3 | 1.0000 | 0.9231 | 0.9600 | 4 | 0.5750 | 0.7931 | 0.6667 | 5 | 0.9565 | 0.8148 | 0.8800 |
|
Table 4. Classification performances of different models
Network | 1D-CNN | 2D-CNN (T-F matrix) | 2D-CNN RGB image |
---|
C1 | (1×5+1)×32=192 | (5×5+1)×32=832 | (1+5×5)×32=832 | C2 | (1+5)×64=384 | (5×5+1)×64=1664 | (1+5×5)×64=1664 | C3 | (1+5)×96=576 | (5×5+1)×96=2496 | (1+5×5)×96=2496 | FC1 | 64×96×1000=614400 | 2×16×96×1000=3072000 | 13×16×96×1000=19968000 | FC2 | 1000×1000=1000000 | 1000×1000=1000000 | 1000×1000=1000000 | Total number of parameters | about 16000 | about 40800 | about 20000 |
|
Table 5. Parameters of CNN with different dimensions
Distance /m | Error accuracy(±1 m)/% | Error accuracy(±2 m)/% | Threat level | Accuracy rate /% |
---|
0 | 100 | 100 | Ⅰ | 100 | 1 | 100 | 100 | 2 | 100 | 100 | 3 | 100 | 100 | 4 | 100 | 100 | 5 | 100 | 100 | Ⅱ | 90.8 | 6 | 100 | 100 | 7 | 100 | 100 | 8 | 71 | 71 | 9 | 87 | 100 | 10 | 83 | 100 | 11 | 100 | 100 | Ⅲ | 100 | 12 | 100 | 100 | 13 | 100 | 100 | 14 | 100 | 100 | 15 | 89 | 100 |
|
Table 6. Model recognition results of the mechanical knock events
[46] Distance/m | Error accuracy(±1 m)/% | Error accuracy(±2 m)/% | Threat level | Accuracy rate/% |
---|
2 | 86 | 86 | Ⅰ | 95.3 | 3 | 100 | 100 | 4 | 100 | 100 | 6 | 33 | 66 | Ⅱ | 65.8 | 8 | 100 | 100 | 11 | 80 | 80 | Ⅲ | 85 | 13 | 100 | 100 | 15 | 80 | 100 | 17 | 60 | 60 |
|
Table 7. Location results of the mechanical excavation
[46]