← Back to Sessions

Using Unsupervised Deep Learning to Denoise Data From Distributed Acoustic Sensing

Description: 

Distributed Acoustic Sensing (DAS) has shown promise in supplementing recordings of traditional sensors. DAS has the advantage that it can be deployed in regions that are difficult to access with conventional sensors, thereby improving sensor coverage for seismic event monitoring. However, unlike seismic stations with well-developed installation standards, DAS cables are typically not well coupled to surrounding rock. This leads to a high noise environment that can vary spatially along the cable. We are adapting an unsupervised deep learning algorithm termed “recorrupted-to-recorrupted” approach (R2R, Pang et al., 2021), which was initially developed for image processing, to train a machine learning denoising model for DAS data. The model is trained using teleseismic and local data recorded along the road section of the Sand Hill dark DAS fiber cable beneath Palo Alto, CA. Preliminary results of the evaluation of the denoising model will be presented.

Session: Advancements in Forensic Seismology and Explosion Monitoring [Poster]

Type: Poster

Date: 4/17/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: James

Student Presenter: No

Invited Presentation: 

Poster Number: 87

Authors