Research On Pattern Recognition Method For Distributed Optical Fiber Vibration Sensing System

With the continuous development of distributed optical fiber vibration sensing technology(DOFVS),its application in perimeter security system has stimulated the new demand of vibration signal pattern recognition to adapt to the the intelligent age.Currently,while most distributed optical fiber vibration monitoring systems used in the complicated outdoor environment encounter a problem that the false response rate is too high,vibration signal pattern recognition technology has a very important research value to solve this problem and provides further improvement to the practicality of DOFVS technology.This thesis takes deep study into the key technologies of pattern recognition method for DOFVS system.Firstly,principle and characteristics of white light interferometric DOFVS system based on linear Sagnac structure are briefly introduced.Structure of the vibration signal sensing units and related settings of signal acquisition are described.Through preprocessing of collected signal,the effects of digital filter denoising,wavelet threshold denoising and signal segmentation algorithm are analyzed.Secondly,core of pattern recognition mechanism is introduced,namely feature extraction and classification algorithm.Suitable feature vector is selected after analysis of the time domain and time-frequency characteristics of five types of vibration signals.The performance of radial basis function(RBF)neural network and support vector machine(SVM)classification algorithm are compared and analyzed,and the best classification strategy is selected.We use MATLAB to do off-line processing with the collected five types of vibration data to verify the feasibility of the overall pattern recognition scheme.Thirdly,C++ implementation of the pattern recognition scheme is presented.Function,class design and key algorithms of 4 modules of the program are elaborated and the interface and function of the software system are briefly described.Finally,experiments on indoor and outdoor real-time vibration type identification of this pattern recognition method are investigated.The field test demonstrates that 5 types of indoor vibration including footsteps of passers-by,bicycle rolling over,rainfall,knocking on the fence and cutting fiber can be correctly recognized at the average accuracy of 92.86%,while 4 types of outdoor vibration including footsteps of passers-by,bicycle rolling over,knocking on the fence and cutting fiber can also be correctly recognized at the average accuracy of 98.1%,which achives significant improvement on the recognition accuracy than existing distributed fiber vibration sensing systems.In grneral,the pattern recognition method of DOFVS system has a high recognition accuracy rate for typical intrusion types,which is very important to improve and strengthen the practicability of DOFVS system.