Multi-Dimensional Distributed Optical Fiber Vibration Sensing Pattern Recognition Based on Convolutional Neural Network

Xibo Jin; Kun Liu; Junfeng Jiang; Shuang Wang; Tianhua Xu; Yuelang Huang; Xinxin Hu; Dongqi Zhang; Tiegen Liu

doi:10.3788/AOS230944

Acta Optica Sinica, Volume. 44, Issue 1, 0106023(2024)

Multi-Dimensional Distributed Optical Fiber Vibration Sensing Pattern Recognition Based on Convolutional Neural Network

Xibo Jin^1,2,3, Kun Liu^1,2,3、*, Junfeng Jiang^1,2,3, Shuang Wang^1,2,3, Tianhua Xu^1,2,3, Yuelang Huang^1,2,3, Xinxin Hu^1,2,3, Dongqi Zhang^1,2,3, and Tiegen Liu^1,2,3

Author Affiliations

¹School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China

²Key Laboratory of Opto-Electronics Information Technology, Ministry of Education, Tianjin University, Tianjin 300072, China

³Institute of Optical Fiber Sensing, Tianjin University, Tianjin 300072, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Schematic diagram of DMZI system

Download full size

Fig. 2. Workflow of unmanned aerial vehicle system

Download full size

Fig. 3. Knocking hard signals. (a) 1D time domain optical signal; (b) 2D time-frequency image after STFT; (c) 2D time-frequency image after STFT with power segmentation

Download full size

Fig. 4. Resnet 50 model. (a) Total structure of optical signal recognition; (b) distribution of convolution layers; (c) architecture of Bottleneck

Download full size

Fig. 5. Schematic of SlowFast model structure

Download full size

Table 1. Parameters of each convolution layers in Resnet 50
View table
Table 1. Parameters of each convolution layers in Resnet 50
Layer Parameters $\{H \times W, C\}$
Conv1 $7 \times 7,64$
Layer 1 $[\begin{array}{l} 1 \times 1,64 \\ 3 \times 3,64 \\ 1 \times 1,256 \end{array}] \times 3$
Layer 2 $[\begin{array}{l} 1 \times 1,128 \\ 3 \times 3,128 \\ 1 \times 1,512 \end{array}] \times 4$
Layer 3 $[\begin{array}{l} 1 \times 1,256 \\ 3 \times 3,256 \\ 1 \times 1,1024 \end{array}] \times 6$
Layer 4 $[\begin{array}{l} 1 \times 1,512 \\ 3 \times 3,512 \\ 1 \times 1,2048 \end{array}] \times 3$

Table 2. Parameters of each convolution layers in SlowFast with Resnet3D-50 backbone

View table

Table 2. Parameters of each convolution layers in SlowFast with Resnet3D-50 backbone

Layer	Parameters $\{T \times S^{2}, C\}$
Layer	Slow pathway	Fast pathway
Data layer	16， $1^{2}$	2， $1^{2}$
Conv1	$1 \times 7^{2}, 64$	$5 \times 7^{2}, 8$
Layer 1	$[\begin{array}{l} 1 \times 1^{2}, 64 \\ 1 \times 3^{2}, 64 \\ 1 \times 1^{2}, 256 \end{array}] \times 3$	$[\begin{array}{l} 3 \times 1^{2}, 8 \\ 1 \times 3^{2}, 8 \\ 1 \times 1^{2}, 32 \end{array}] \times 3$
Layer 2	$[\begin{array}{l} 1 \times 1^{2}, 128 \\ 1 \times 3^{2}, 128 \\ 1 \times 1^{2}, 512 \end{array}] \times 4$	$[\begin{array}{l} 3 \times 1^{2}, 16 \\ 1 \times 3^{2}, 16 \\ 1 \times 1^{2}, 64 \end{array}] \times 4$
Layer 3	$[\begin{array}{l} 3 \times 1^{2}, 256 \\ 1 \times 3^{2}, 256 \\ 1 \times 1^{2}, 1024 \end{array}] \times 6$	$[\begin{array}{l} 3 \times 1^{2}, 32 \\ 1 \times 3^{2}, 32 \\ 1 \times 1^{2}, 128 \end{array}] \times 6$
Layer 4	$[\begin{array}{l} 3 \times 1^{2}, 512 \\ 1 \times 3^{2}, 512 \\ 1 \times 1^{2}, 2048 \end{array}] \times 3$	$[\begin{array}{l} 3 \times 1^{2}, 64 \\ 1 \times 3^{2}, 64 \\ 1 \times 1^{2}, 256 \end{array}] \times 3$

Table 3. Data distribution of optical signals and video signals

View table

Table 3. Data distribution of optical signals and video signals

Label	Intrusion event	Optical signal			Video signal
Label	Intrusion event	Train	Validation	Test	Train	Validation	Test
1	Climbing	1440	180	180	112	14	14
2	Crashing	1440	180	180	112	14	14
3	Cutting	1440	180	180	112	14	14
4	Kicking	1440	180	180	112	14	14
5	Knocking hard	1440	180	180	112	14	14
6	Knocking lightly	1440	180	180	112	14	14
7	Pulling	1440	180	180	112	14	14
8	Waggling	1440	180	180	112	14	14
9	No intrusion	1440	180	180	0	0	0

Table 4. Comparison of recognition results of 3 different methods

View table

Table 4. Comparison of recognition results of 3 different methods

Method	Event type	P /%	R /%	F₁ /%
Using 2D time-frequency images （A=96.22%）	Label 1	100.00	100.00	100.00
	Label 2	88.00	100.00	93.62
	Label 3	100.00	100.00	100.00
	Label 4	90.00	100.00	94.74
	Label 5	100.00	86.21	92.59
	Label 6	99.00	100.00	99.50
	Label 7	100.00	98.04	99.01
	Label 8	93.00	92.59	92.80
	Label 9	100.00	92.59	96.15
Using 3D video signals （A=78.59%）	Label 1	30.35	22.91	26.11
	Label 2	100.00	100.00	100.00
	Label 3	100.00	82.42	90.36
	Label 4	100.00	100.00	100.00
	Label 5	23.68	100.00	38.29
	Label 6	100.00	100.00	100.00
	Label 7	75.26	60.79	67.26
	Label 8	100.00	100.00	100.00
Proposed method （A=99.58%）	Label 1	100.00	100.00	100.00
	Label 2	100.00	100.00	100.00
	Label 3	100.00	100.00	100.00
	Label 4	100.00	100.00	100.00
	Label 5	100.00	100.00	100.00
	Label 6	100.00	100.00	100.00
	Label 7	100.00	100.00	100.00
	Label 8	100.00	100.00	100.00
	Label 9	100.00	92.59	96.15

Table 5. Comparison of different pattern recognition methods
View table
Table 5. Comparison of different pattern recognition methods
Model Response time /s
Number of
intrusion events
Prediction
average accuracy /%
SVM-RBF^［4］ 1.01 5 97.10
CLDNN^［5］ 4.00 3 97.00
5 layers CNN^［6］ 5.00 6 93.47
YOLOv5s^［7］ 0.30 5 96.60
Proposed method 0.30 9 99.58

Tools

Get Citation

Copy Citation Text

Xibo Jin, Kun Liu, Junfeng Jiang, Shuang Wang, Tianhua Xu, Yuelang Huang, Xinxin Hu, Dongqi Zhang, Tiegen Liu. Multi-Dimensional Distributed Optical Fiber Vibration Sensing Pattern Recognition Based on Convolutional Neural Network[J]. Acta Optica Sinica, 2024, 44(1): 0106023

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites