Optical Fiber Perimeter Intrusion Event Recognition Based on ISSA and Genetic Algorithm Optimized BiLSTM Neural Network

Yuzhao Ma; Tingting Zhang; Qingxiao Zhu; Meng Li

doi:10.3788/LOP230695

Laser & Optoelectronics Progress, Volume. 61, Issue 5, 0506001(2024)

Optical Fiber Perimeter Intrusion Event Recognition Based on ISSA and Genetic Algorithm Optimized BiLSTM Neural Network

Yuzhao Ma^1、*, Tingting Zhang¹, Qingxiao Zhu¹, and Meng Li²

Author Affiliations

¹College of Electronic Information and Automation, Civil Aviation University of China, Tianjin 300300, China

²College of Air Traffic Management, Civil Aviation University of China, Tianjin 300300, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(21)

Fig. 1. Fiber perimeter system based on dual Mach-Zehnder

Download full size

Fig. 2. Schematic diagram of ISSA method

Download full size

Fig. 3. BiLSTM neural network structure diagram

Download full size

Fig. 4. Principle diagram of GA-BiLSTM neural network recognize intrusion events

Download full size

Fig. 5. Fiber perimeter intrusion signal. (a) Climb; (b) run; (c) knock; (d) static; (e) wind; (f) rain

Download full size

Fig. 6. Knock signal singular spectrum analysis. (a) Distribution of singular value; (b) distribution of contribution rate

Download full size

Fig. 7. Time-frequency analysis of knock signal. (a) Time domain; (b) frequency domain

Download full size

Fig. 8. Number of main signal frequencies

Download full size

Fig. 9. Time-frequency analysis of knock signal components after one SSA. (a) Time domain;(b) frequency domain

Download full size

Fig. 10. Time-frequency analysis of knock signal components after twice SSAs. (a) Time domain; (b) frequency domain

Download full size

Fig. 11. Time-frequency analysis of knock signal components after three SSAs. (a) Time domain;(b) frequency domain

Download full size

Fig. 12. Recognition accuracy of optimization algorithms under different evolutions. (a) 24 times; (b) 72 times; (c) 120 times; (d) 240 times

Download full size

Fig. 13. Fitness change curve

Download full size

Fig. 14. RNN recognition effect diagram

Download full size

Fig. 15. GA-RNN recognition effect diagram

Download full size

Fig. 16. BiLSTM neural network recognition effect diagram

Download full size

Fig. 17. GA-BiLSTM neural network recognition effect diagram

Download full size

Table 1. Laser parameters
View table
Table 1. Laser parameters
Wavelength /nm Power /mW Line width /kHz
1550 20 10

Table 2. Comparison of denoising performance of SSA and ISSA

View table

Table 2. Comparison of denoising performance of SSA and ISSA

Denoising method	Signal type	Average signal-to-noise ratio /dB	Mean square error
SSA	Climb	38.6304	0.00160
	Run	36.2612	0.00190
	Knock	30.1120	0.00210
	Static	42.0467	0.00052
	Wind	40.4972	0.00140
	Rain	47.3670	0.00054
ISSA	Climb	50.8602	0.00120
	Run	47.1619	0.00160
	Knock	38.8160	0.00230
	Static	58.8285	0.00027
	Wind	53.9244	0.00081
	Rain	62.0710	0.00018

Table 3. Neural networks recognition time for different evolutions
View table
Table 3. Neural networks recognition time for different evolutions
Recognition method Evolved 24 times Evolved 72 times Evolved 120 times Evolved 240 times
GA-RNN 1.62 5.66 8.37 15.83
GA-BiLSTM 1.38 4.14 5.68 10.44

Table 4. Comparison of recognition effect of different neural networks

View table

Table 4. Comparison of recognition effect of different neural networks

Signal type	RNN	GA-RNN	BiLSTM	GA-BiLSTM
Climb	78.3	84.3	89.4	96.9
Run	56.1	83.6	94.3	98.8
Knock	52.5	80.2	91.7	99.8
Static	76.0	89.6	92.1	99.2
Wind	65.6	81.7	93.8	98.3
Rainy	73.7	89.5	95.1	95.3
Average recognition rate /%	67.0	84.8	92.8	98.1

Tools

Get Citation

Copy Citation Text

Yuzhao Ma, Tingting Zhang, Qingxiao Zhu, Meng Li. Optical Fiber Perimeter Intrusion Event Recognition Based on ISSA and Genetic Algorithm Optimized BiLSTM Neural Network[J]. Laser & Optoelectronics Progress, 2024, 61(5): 0506001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites