Study On Distributed Optical Fiber Security Detection System Signal Recognition

With the rapid development of society, the security monitoring issue of communication cables, pipelines, gas pipelines, high-voltage power grid and military, civilian security systems and other infrastructure equipment has increasingly become an important factor affecting economic development and social stability. So it has important research value and practical significance to discover timely and position effectively all kinds of illegal intrusions against the safety of cables, pipelines and security systems.This subject originates from the national 973 project of cooperation with Tianjin University-Basic Research of Key Technology of Intelligent Optical Fiber Sensor Network Experimental Platform and Its Application. And it focuses on the subproject Study on Adverse Environmental Effects on Polarization and Positioning of Continuously Distributed Sensor Networks. The paper aims at improving the security system disturbance event recognition accuracy of optical fiber Mach-Zehnder interferometer by means of wavelet multi-resolution analysis and neural networks to make signal feature extraction and recognition on collected monitoring data of distributed optical fiber sensing security systems. And it achieved the recognition to external environmental signals from rain, percussion, climbing, which provided the research foundation on positioning accurately the intrusion in security monitoring system.This paper mainly completed the following work. (1) Realized multiresolution information extraction on the monitoring data of distributed optical fiber sensing security systems by using wavelet transform. Realized vector extraction of signal feature through the wavelet energy spectrum. And it achieved the essential characteristics better reflecting the various security incidents and the initial recognition on the signals of rain, percussion, climbing and so on. (2) Understand depth on the neural network architecture, applies the BP neural network to recognizing the types of the detection signals. Designed for four improved BP neural network in MATLAB program and use the obtained characteristics of the sample data to train these neural network make a comparison of these training results, select the best training method of BP network. (3) Monitoring data collected in the field experiments conducted experimental studies to prove the validity and reliability of the algorithm, recognition rate of BP neural network up to 90%.