A regular composite feature extraction method for vibration signal pattern recognition in optical fiber link system

Abstract

Phase optical time-domain reflectometer link monitoring system is a kind of sensor system which uses optical fiber as sensing medium. It can detect, recognize and locate invasive signals. The module of pattern recognition, which is one of the important parts of the system, can intelligently and instantly distinguish dangerous intrusions from safe disturbances. This paper proposes a regular composite feature extraction method which can be used for vibrational signal pattern recognition in an optical fiber monitoring system. This method applies an improved double-threshold method to detect the start-stop positions of the valid signal segments, and then combines the maximum energy and maximum signal noise ratio to select the main invasion feature segment within a sampling period. The composite eigenvectors, which are extracted by using the last time of the feature segment and wavelet packet energy spectrum, can be used by the support vector machine to recognize the signal patterns. Experiment results show that the accuracy rate of the pattern recognition has been improved significantly based on the proposed method in this paper.