What is it about?
How to use deep learning to attenuate noise without labelled data? For addressing the similar issue, namely the shear wave noise attenuation from OBN 4C data, this paper introduces a CNN-based workflow to generate the training set without labelled noise-free z-component data and attenuate the noise by using the pretrained network.
Featured Image
Photo by Thierry Meier on Unsplash
Why is it important?
Our examples show that the training set generated by adding the field x-component data to p-component data is representative enough for attenuating the noisy z-component data. Thus, the labelled noise-free z-component data is unnecessary for generating the training set. It indicates that we do not have to use the target labelled data to generate the training set.
Perspectives
We have been thinking over such a question how to use supervised CNN methods to denoise seismic data without field noise-free data. This article provides us a good example for addressing this issue. We need to ensure that the training set and test set share similar noise features and signal textures, rather than improving the SNR of the field data through preprocessing.
Shaowen Wang
Ocean University of China
Read the Original
This page is a summary of: Deep learning-based attenuation for shear wave leakage from ocean-bottom node data, Geophysics, January 2023, Society of Exploration Geophysicists,
DOI: 10.1190/geo2022-0252.1.
You can read the full text:
Contributors
The following have contributed to this page
