Research On Data Processing And Target Recognition Algorithm Of Optical Fiber Sensing Based On BP Neural Network Model

Optical fiber vibration sensing system has high application value in the field of safety monitoring because of its excellent characteristics such as lightweight,high temperature resistance,corrosion resistance and strong electromagnetic interference resistance.Phasesensitive Optical Time Domain Reflectometer(Φ-OTDR)is widely used in long-distance distributed vibration monitoring because of its high sensitivity,simple structure and simultaneous measurement of multiple points.The vibration sensing signals of intrusion behavior in outdoor environment are very complex,it is of great research value to realize accurate target recognition of intrusion behavior.In this thesis,the target recognition algorithm of optical fiber vibration sensing system is deeply studied,the collected vibration sensing signals are processed and analyzed,the recognition model is established and the program is written.The main research contents are as follows:(1)Based on the principles and technologies of Rayleigh scattering,Φ-OTDR and machine learning,the target recognition algorithm of distributed optical fiber vibration sensing system is studied;(2)Five kinds of different vibration sensing signals,such as no interference,shaking the fence,cutting the optical cable with a knife,hammering and rain washing,are collected through experiments.The sensing signals are preprocessed by wavelet threshold denoising and fragment segmentation,the time-frequency-domain features and time-domain features are extracted,the features are selected and fused,and the feature data sets of five kinds of vibration sensing signals are obtained;(3)The Back Propagation(BP)neural network is used to train the recognition model of the feature data sets,the average recognition accuracies obtained from other feature extraction methods are analyzed and compared,and it is concluded that the average recognition accuracy has been improved by at least 0.90%.By optimizing the BP neural network,the final recognition model is tested and its performance evaluation indexes are obtained.Compared with other traditional machine learning algorithms,the average recognition accuracy of the designed vibration sensing signal recognition model has been improved by at least 1.60%.Finally,the target recognition program of vibration sensing signal is designed and written.The experimental results show that the feature extraction method of vibration sensing signal proposed in this thesis can achieve the average recognition accuracy of 98.40% for five types of vibration sensing signals through the optimized BP neural network recognition model,and the performance evaluation indexes of the optimized BP neural network recognition model are all excellent,which meets the requirements of accurate recognition.