Research On Pattern Recognition Of Distributed Optical Fiber Vibration Sensing For Safety Monitoring

Safety monitoring,as a prerequisite for ensuring the stable and sustained production of people's lives,has attracted more and more attention in today's rapid economic development.Optical fiber sensing technology transforms external information into visual data through fiber optic links,and has been widely used in security monitoring for a long time,as an important branch of optical fiber sensing technology,distributed optical fiber sensing technology is more conducive to the formation of security monitoring sensor network due to the infinite extension of the link axis.At the same time,in order to cope with the high false positive rate of safety monitoring in complex environments,the pattern recognition discipline has developed rapidly with the rise of artificial intelligence which has important research value in the application of distributed optical fiber sensing technology in security monitoring.In view of the above security monitoring requirements,we studied two types of distributed optical fiber vibration sensors based on white light interference and phase sensitive optical time domain reflection,and focused on experimental verification of two typical applications including fiber perimeter safety monitoring and pipeline health monitoring.At the same time,we performed feature analysis and pattern recognition on the interference events of the two types of applications,and finally realized the accurate identification of various event types.The main contents of the thesis include:(1)A distributed optical fiber sensor based on a white light interference type linear Sagnac structure has been studied,and we set up the corresponding experimental platform for event simulation and vibration signal acquisition of four typical interference events in fiber perimeter safety monitoring(pedestrian walking,bicycles passing,knocking fences and cutting fibers),then pre-processed the signals of these four types of interference events including wavelet threshold denoising and signal segmentation.(2)A pattern recognition algorithm based on support vector machine(SVM)for four kinds of interference events of fiber perimeter security monitoring is proposed.We performed feature extraction and analysis on the segmentation signals of four kinds of interference events including(time domain)segment interval,segment length,segment intensity peak-to-average ratio,and(frequency domain)band energy distribution;The SVM method is used for classifier training and pattern recognition for four kinds of interference events in fiber perimeter security monitoring,and We finally achieved a higher perimeter interference event recognition accuracy rate.(3)A distributed optical fiber sensor based on phase-sensitive optical time domain reflectometry(?-OTDR)has been studied.An experimental platform is set up to perform event simulation,vibration signal detection and pre-processing on pipeline damage detection and pipeline emergency warning(pipe leakage,third-party excavation)in pipeline health monitoring.(4)A pattern recognition algorithm based on artificial neural network(ANN)for pipeline health monitoring is proposed.We performed separate time,frequency domain analysis and feature extraction for pipeline damage areas in damage detection,and leakage and third-party mining in emergency warning;The method of ANN is used to classifier training and pattern recognition of the whole damaged pipeline and specific damaged area in pipeline damage detection,and we finally achieved damage detection of damaged pipes;At the same time,we use the same method to perform classifier training and pattern recognition for leaks and third-party mining of pipeline incident warnings,and finally realized the event detection of two kinds of interference events.