Identification Method of Optical Fiber Perimeter Intrusion Signal Based on MATCN

Qiufeng Shang; Da Huang

doi:10.3788/AOS230873

Acta Optica Sinica, Volume. 43, Issue 20, 2006006(2023)

Identification Method of Optical Fiber Perimeter Intrusion Signal Based on MATCN

Qiufeng Shang^1,2,3 and Da Huang^1、*

¹Department of Electronic & Communication Engineering, North China Electric Power University, Baoding 071003, Hebei, China

²Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, Hebei, China

³Baoding Key Laboratory of Optical Fiber Sensing and Optical Communication Technology, North China Electric Power University, Baoding 071003, Hebei, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(17)

Fig. 1. Schematic of causal convolution

Download full size

Fig. 2. Schematic of dilated causal convolution

Download full size

Fig. 3. Residual module in different networks. (a) TCN; (b) MATCN

Download full size

Fig. 4. Comparison of different activation units. (a) ReLU; (b) Leaky ReLU

Download full size

Fig. 5. Channel attention mechanism

Download full size

Fig. 6. Temporal attention mechanism

Download full size

Fig. 7. MATCN model

Download full size

Fig. 8. Experimental system structure

Download full size

Fig. 9. Vibration event location and typical waveforms. (a) Climbing; (b) knocking; (c) trampling; (d) no invasion

Download full size

Fig. 10. Change of loss function values of residual modules at different levels

Download full size

Fig. 11. t-SNE visualization result. (a) Characteristic distribution of the original signal; (b) feature extraction result

Download full size

Fig. 12. Training curve of each network structure

Download full size

Fig. 13. Recognition rate and loss value curves of different networks for validation samples. (a) LSTM; (b) CNN-LSTM; (c) TCN;(d) MATCN

Download full size

Fig. 14. Recognition result comparison of machine learning algorithms and MATCN

Download full size

Table 1. Identification results of each network structure for the validation sample
View table
Table 1. Identification results of each network structure for the validation sample
Parameter Net1 Net2 Net3 Net4
Accuracy /% 96.56 96.87 98.82 99.03
Loss 0.051 0.033 0.011 0.005

Table 2. Comparison of training efficiency of four networks
View table
Table 2. Comparison of training efficiency of four networks
Network Iteration time /（ms·step^-1） One training epoch /s Number of epochs Training time /s
LSTM 68 21.76 23 501
CNN-LSTM 74 23.68 32 758
TCN 14 4.48 29 130
MATCN 19 6.08 15 91

Table 3. Recognition rate and efficiency of four networks for non-training samples

View table

Table 3. Recognition rate and efficiency of four networks for non-training samples

Network	Accuracy of climbing /%	Accuracy of knocking /%	Accuracy of trampling /%	Accuracy of no invasion /%	Average accuracy /%	Testing time /s
LSTM	93.40	88.40	81.60	94.40	89.45	2.03
CNN-LSTM	98.00	92.20	87.20	99.20	94.10	2.10
TCN	96.60	93.40	86.20	97.20	93.04	0.39
MATCN	99.60	98.80	96.80	98.80	98.50	0.53

Tools

Get Citation

Copy Citation Text

Qiufeng Shang, Da Huang. Identification Method of Optical Fiber Perimeter Intrusion Signal Based on MATCN[J]. Acta Optica Sinica, 2023, 43(20): 2006006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Fiber Optics and Optical Communications

Received: Apr. 25, 2023

Accepted: May. 29, 2023

Published Online: Oct. 13, 2023

The Author Email: Huang Da (huangda0217@163.com)

DOI:10.3788/AOS230873

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Identification results of each network structure for the validation sample

Table 1. Identification results of each network structure for the validation sample

Table 2. Comparison of training efficiency of four networks

Table 2. Comparison of training efficiency of four networks

Table 3. Recognition rate and efficiency of four networks for non-training samples

Table 3. Recognition rate and efficiency of four networks for non-training samples