A Self-Supervised Deep Learning Approach for Blind Denoising and Waveform Coherence Enhancement in Distributed Acoustic Sensing Data

Ende, Martijn van den; Lior, Itzhak; Ampuero, Jean‐Paul; Sladen, Anthony; André, Fabrice; Richard, Cédric

doi:10.31223/x55k63

Cited by 8 publications

(4 citation statements)

References 44 publications

(9 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(2021); and in van den Ende et al. (2021), similar concept was shown for distributed acoustic sensing (DAS). Aftershock analysis . Large earthquakes are often accompanied by many aftershocks (Ross et al., 2018), and their number usually decays exponentially (Baranov et al., 2019).…”

Section: Introductionmentioning

confidence: 69%

See 1 more Smart Citation

SEDENOSS: SEparating and DENOising Seismic Signals With Dual‐Path Recurrent Neural Network Architecture

2022

View full text Add to dashboard Cite

Seismologists have to deal with overlapping and noisy signals. Techniques such as source separation can be used to solve this problem. Over the past few decades, signal processing techniques used for source separation have advanced significantly for multi‐station settings. But not so many options are available when it comes to single‐station data. Using Machine Learning, we demonstrate the possibility of separating signals for single‐station, one‐component seismic recordings. The technique that we use for seismic signal separation is based on a dual‐path recurrent neural network which is applied directly to the time domain data. Such source separation may find applications in most tasks of seismology, including earthquake analysis, aftershocks, nuclear verification, seismo‐acoustics, and ambient‐noise tomography. We train the network on seismic data from STanford EArthquake Dataset and demonstrate that our approach is (a) capable of denoising seismic data and (b) capable of separating two earthquake signals from one another. In this work, we show that Machine Learning is useful for earthquake‐induced source separation. We provide a reproducible research repository with the algorithms here: https://github.com/crimeacs/source-separation.

show abstract

Section: Introductionmentioning

confidence: 69%

“…(2021); and in van den Ende et al. (2021), similar concept was shown for distributed acoustic sensing (DAS).…”

Section: Introductionmentioning

confidence: 69%

SEDENOSS: SEparating and DENOising Seismic Signals With Dual‐Path Recurrent Neural Network Architecture

2022

View full text Add to dashboard Cite

show abstract

“…More recently, self-supervised denoising algorithms based on machine leaning and, more specifically, on deep learning approaches have been proposed [60], [61]. These blind-denoising methods typically achieve a higher level of signal to noise ratio than linear filtering when removing spatio-temporally incoherent noise, with no need to provide noise-free ground truth.…”

Section: B Denoising Algorithmsmentioning

confidence: 99%

Seismic Monitoring With Distributed Acoustic Sensing From the Near-Surface to the Deep Oceans

Fernández‐Ruiz

Martins

Williams

et al. 2022

J. Lightwave Technol.

View full text Add to dashboard Cite

Distributed acoustic sensing (DAS) delivers real-time observation of physical perturbations such as vibrations or strain variations in conventional optical fibers with high sensitivity. The high density of sensing points and large network footprint provided by a single DAS system, along with the availability of a vast optical fiber network already deployed both in land and in oceanic regions, contrast with the high deployment and maintenance cost of conventional instrumentation networks for seismology. This situation has triggered a rapid growth of DAS deployments for seismic monitoring in recent years. Photonic engineers and geophysicists have joined efforts to prove the value of optical fibers as distributed seismometers, which has resulted in a wide panoply of tests demonstrating diverse applicability across the geosciences. For example, DAS has been successfully applied recording local to teleseismic earthquakes, monitoring glacial icequakes, and observing oceanographic phenomena at the sea floor. Most of the realized tests have been performed using commercially available optical fiber interrogators based on phase-sensitive optical time-domain reflectometry. Among them, DAS based on chirped pulse distributed Manuscript

show abstract

“…We build on this approach in this study. More recently, van den Ende et al (2021) proposed a deep learning approach for removing incoherent signal with a focus on DAS data.…”

mentioning

confidence: 99%

A unified wavefield-partitioning approach for distributed acoustic sensing

Atterholt

Zhan

Shen

et al. 2021

Geophysical Journal International

View full text Add to dashboard Cite

Summary While Distributed Acoustic Sensing (DAS) has been demonstrated to have great potential in seismology, DAS data often have much higher levels of stochastic and coherent noise (e.g., instrument noise, traffic vibrations) than data collected by traditional seismometers. The linearly, densely spaced nature of DAS arrays presents a suite of opportunities for more innovative processing techniques that can be used to address this issue. One way to take advantage of DAS’s array architecture is through the use of curvelets. Curvelets have a non-uniform scaling property that makes them an excellent tool for representing images with discontinuities along piecewise, twice continuously differentiable curves. This anisotropic scaling property makes curvelets an ideal processing tool for DAS data, for which the measured wavefield can be represented as an image composed of curved features. Here we use the curvelet frame as a tool for the manipulation of DAS signal, and we demonstrate how this manipulation can improve our ability to identify important features in DAS datasets. We use the curvelet representation to partition the measured wavefield using DAS data collected near Ridgecrest, CA following the 2019 Mw7.1 Ridgecrest earthquake. Here we isolate the earthquake induced wavefield from coherent and stochastic noise using the curvelet frame in an effort to improve the results of template matching of the Ridgecrest aftershock sequence. We show that our wavefield partitioning technique facilitates the identification of over 30 per cent more aftershocks and greatly reduces the magnitude of diurnal depressions in the aftershock catalog due to cultural noise.

show abstract

A Self-Supervised Deep Learning Approach for Blind Denoising and Waveform Coherence Enhancement in Distributed Acoustic Sensing Data

Cited by 8 publications

References 44 publications

SEDENOSS: SEparating and DENOising Seismic Signals With Dual‐Path Recurrent Neural Network Architecture

SEDENOSS: SEparating and DENOising Seismic Signals With Dual‐Path Recurrent Neural Network Architecture

Seismic Monitoring With Distributed Acoustic Sensing From the Near-Surface to the Deep Oceans

A unified wavefield-partitioning approach for distributed acoustic sensing

Contact Info

Product

Resources

About