Research On Pattern Recognition Of Vibration Signal In ?-OTDR Optical Fiber Sensing System

Phase-sensitive Optical Time Domain Reflectometer(?-OTDR)can realize continuous distributed measurement and multi-point positioning for vibration signal with high sensitivity,and has high application value in perimeter security,oil and gas pipeline leakage monitoring and other fields.Compared with point measurement,?-OTDR can realize distributed remote measurement with high resolution and measurement sensitivity.Based on these advantages,the research of ?-OTDR optical fiber sensor system has become one of the main research directions of optical fiber sensor monitoring system,which mainly includes the research of demodulation method of sensor signal,location method of disturbance event and pattern recognition of vibration signal.The identification technology of vibration signal is an important part of the monitoring system,which can effectively distinguish interference events and harmful intrusion events,reduce false and missed reports,timely feedback accurate status information,and help the monitoring personnel to make countermeasures.In this paper,the problem of pattern recognition of vibration signal in ?-OTDR optical fiber sensing system is studied.A pattern recognition algorithm based on two-level feature selection is proposed,which integrates one-dimensional signal feature and two-dimensional image feature.The vibration signal in ?-OTDR optical fiber sensing system is processed from two angles of one-dimensional signal andtwo-dimensional image,and the fusion feature is extracted,and the support vector machine(SVM)is used for pattern recognition.In the aspect of signal preprocessing,for one-dimensional signal,the method of wavelet packet decomposition is used to remove the low-frequency trend term interference,and then the method of wavelet packet threshold denoising is used for noise reduction.Finally,based on the double threshold method of speech endpoint detection and the characteristics of vibration signal in ?-OTDR optical fiber sensing system,the short-time energy of each frame of signal is calculated to extract the effective segments of vibration signal.For the spatiotemporal two-dimensional image,the short-time variance of the spatiotemporal two-dimensional signal collected by the?-OTDR optical fiber sensor system is processed to obtain the short-time variance spatiotemporal two-dimensional image,and then the feature area is extracted by the method of binary image connected domain marking.The image processing includes gray-scale transformation,closed operation,binarization,connected domain marking and small noise removal.In the aspect of feature extraction,the time-domain,time-frequency-domain,singular value,singular spectrum features of the effective segments of vibration signal and the morphological features of the two-dimensional image feature areas are extracted to form the initial feature set.In the aspect of feature selection,the two-level feature selection method combining Fisher criterion and genetic algorithm is used to select feature subsets with high classification accuracy and small feature size.The feature vector corresponding to the selected feature subset is input into SVM classifier for training,and the final SVM model of ?-OTDR vibration signal pattern recognition is obtained.In the laboratory environment,five kinds of vibration signals,such as shovel digging,Sandbox knocking,jumping,rubber hammer knocking and speaker vibration,are obtained as experimental data.After verification,the average recognition accuracy of the algorithm is 99.22%.The experimental result shows that for the vibration signal pattern recognition in the ?-OTDR optical fiber sensor system,the pattern recognition algorithm based on two-level feature selection,which integrates one-dimensional signal features and two-dimensional space-time image features,can reduce the size of feature subset,and has high recognition accuracy.