What is it about?

This research introduces a new method to clean up seismic data — the kind of data used to understand underground structures, like for oil exploration or earthquake studies. Normally, computers need examples of “clean” data to learn how to remove noise, but in the real world, such clean data is hard to get. This study shows how we can train a computer to remove noise using only noisy data, by applying a technique called "Noise2Noise". It even works well with real-world data and under challenging conditions, making the process more practical and cost-effective.

Featured Image

Why is it important?

This work solves a long-standing problem in geophysics: how to improve data quality when clean data isn’t available. By eliminating the need for noise-free examples, this method can help reduce the cost and effort of seismic surveys, especially in real-world field conditions. It also opens up new possibilities for long-term underground monitoring, such as tracking CO₂ storage or earthquake activity, in a faster and more efficient way.

Perspectives

As someone working closely with seismic data in practical field settings, I’ve often struggled with the lack of clean data for training machine learning models. This limitation motivated me to explore alternatives that could work directly with noisy field data. Developing and testing the Noise2Noise-enhanced framework in both repeated and single-shot scenarios was personally rewarding, especially when I saw it perform well on real field datasets. I believe this approach brings us a step closer to making deep learning more accessible and realistic in seismic data processing, and I hope it inspires further innovations in the field.

Mitsuyuki Ozawa
JGI, Inc.

Read the Original

This page is a summary of: Enhancing seismic noise suppression using the Noise2Noise framework, Geophysics, November 2024, Society of Exploration Geophysicists,
DOI: 10.1190/geo2024-0106.1.
You can read the full text:

Read

Contributors

The following have contributed to this page