Data-driven classification of events in the vicinity of an optical fibre

Abstract

This article focuses on the data-driven classification of vibration sources via a distributed optical fibre sensor system utilizing Gaussian Mixture Models and Mel-spectrograms. In total, six categories are considered: one person walking/running, group walking/running, hammering, and car riding. Two evaluation scenarios are used: individual models (per-session) and a joint model. Tenfold cross-validation is employed during the evaluation phase.