Efficient Autoencoders for Anomaly Detection in Distributed Acoustic Sensing Data

Distributed Acoustic Sensing (DAS) has become a promising technology for many disciplines, including geophysics and marine biology. However, current processing and anomaly detection methods often prioritize accuracy over efficiency.This work addresses the challenge of efficiently processing DAS data while maintaining effective anomaly detection capabilities. We build a novel framework that explores applying compact autoencoders for rapid anomaly detection of DAS data. Our framework is validated using an open-source dataset. We obtain promising results using a convolutional autoencoder with only 47k parameters that can detect anomalies in large matrices in just 5.6ms while trained on potentially anomalous data. Our novel work focuses on optimizing the combination of high accuracy, low FPR rates, smaller models and low inference times. Our efficient autoencoder design paves the way for the use of DAS technology in real-time scenarios.