Pattern Recognition of Fiber Disturbance Based on Support Vector Machine in Polarization Optical Time Domain Reflectometry

摘要

Aiming at improving the signal processing capability of polarization optical time domain reflectometry (POTDR), arecognition method mainly based on the feature extraction and supported vector machine (SVM) is proposed. Apart fromlocating the certain place of interruptions, this method can help us identify different kinds of intrusion events. Firstly, wepreprocess the signal by using an average filter and setting a proper threshold for it. Secondly, the signal is transformedinto various kinds of time-domain features and frequency-domain features for the subsequent classification. Finally, theSVM of the system is trained with initial signals so it can discriminate events represented by new signal accurately. Ourexperiment results show the effectiveness of this method, and it can work well with high accuracy, fast response speedand low cost.