Classification Studies on Vibrational Patterns of Distributed Fiber Sensors using Machine Learning

Abstract

Distributed fiber optic sensors are smart replacements to point sensors in monitoring vibrations over long distances with excellent resolution. In this paper, we investigate the use of machine learning models to classify different vibrational events. Spectrograms of vibrational events available on a public database is used for training and testing the machine learning models like Support Vector Machine, Ensemble learning and K-Nearest Neighbour. The best accuracy of 86.1% is obtained for Support Vector classifier after hyperparameter tuning with 5-fold cross validation.