Deep Generative Models for Distributed Acoustic Sensors (DAS)

Abstract

This paper presents two solutions based on Deep Learning techniques to detect mechanical events in signals coming from distributed acoustic sensors (DAS). Specifically, two systems for this task are described. The first one is a deterministic solution based on the concept of autoencoder (AE), while the second system is a stochastic solution based on the idea of Variational Autoencoder (VAE). The signals used for the tests have been provided by Aragón Photonics Labs (APL).