Comprehensive Dataset for Event Classification Using Distributed Acoustic Sensing (DAS) Systems

Adrian Tomasov ¹, Pavel Zaviska ², Petr Dejdar ²

⁰Brno University of Technology, FEEC, Dept. of Telecommunications, Technicka 12, 61600, Brno, Czech Republic. tomasov@vutbr.cz.

Scientific Data +

|

May 14, 2025

View abstract on PubMed

Summary

This study introduces a new labeled dataset for Distributed Acoustic Sensing (DAS) events, crucial for improving machine learning models. The dataset enables more accurate classification of acoustic signals from walking, running, and vehicles.

Area Of Science

Geophysics and sensor technology
Machine learning applications
Data science for signal processing

Background

Distributed Acoustic Sensing (DAS) uses optical fibers for high-resolution acoustic detection over long distances.
Accurate event classification in DAS is vital for applications like seismic and structural monitoring, but is challenged by noisy, high-dimensional data.
Existing machine learning approaches are limited by the scarcity of large, high-quality labeled DAS datasets.

Purpose Of The Study

To present a novel, comprehensive, and labeled dataset of Distributed Acoustic Sensing (DAS) measurements.
To facilitate the development and validation of advanced machine learning models for DAS event classification.
To address the critical need for high-quality data in advancing DAS technology.

Main Methods

Collection of DAS measurements across a university campus environment.
Labeling of the dataset to include diverse events such as pedestrian and vehicular movement, and potential security events.
Demonstration of dataset utility through training a Convolutional Neural Network (CNN) model.

Main Results

A valuable, labeled dataset of DAS measurements has been successfully created and curated.
The dataset supports the development of machine learning models for enhanced event classification.
Successful training of a CNN model validates the quality and utility of the presented dataset.

Conclusions

The developed labeled DAS dataset is a significant resource for the research community.
This dataset will accelerate progress in automated and accurate event classification for DAS systems.
The findings highlight the potential of machine learning, powered by quality data, to overcome challenges in DAS signal processing.