Pre‐Trained Foundation Model for Uncorrelated Noise Attenuation of Seismic Data
Geophysical Prospecting · 2026-11 · 74 (9)
DOI: 10.1111/1365-2478.70285 ↗Kurz gefasst
Fachartikel in der Zeitschrift Geophysical Prospecting, online erschienen am 16.11.2026. Verfasst von Chao Fu und 2 weiteren Autor:innen. Die Arbeit nennt 65 Literaturangaben. Gefördert von: Natural Science Foundation of Inner Mongolia Autonomous Region.
Abstract (Originalsprache)
High‐quality seismic data contribute to improving the accuracy of subsequent imaging and inversion. Methods based on artificial intelligence (AI) for uncorrelated noise attenuation have achieved promising performance. Nevertheless, existing AI denoising approaches require retraining for different datasets, leading to excessive computational redundancy and limited generalization ability. Consequently, this study proposes a seismic data noise attenuation method based on foundation models. Using the publicly available seismic datasets from the denoising benchmark, we fed massive seismic data into the network to learn the characteristics of random noise and erratic noise for model training. The trained network exhibits excellent generalization performance. For practical denoising tasks, the network can be rapidly deployed via fine‐tuning with minimal effort. Results from synthetic and field data demonstrate that the proposed method can effectively suppress noise with reduced model parameters and training time. Meanwhile, it exhibits strong generalization ability. For different datasets, favourable performance can be achieved with only a small number of training epochs.
Förderung
- Natural Science Foundation of Inner Mongolia Autonomous Region
Zitieren
APA
Fu, C., Ma, Y., & Xu, X. (2026). Pre‐Trained Foundation Model for Uncorrelated Noise Attenuation of Seismic Data. Geophysical Prospecting, 74(9). https://doi.org/10.1111/1365-2478.70285
Vancouver
Fu C, Ma Y, Xu X. Pre‐Trained Foundation Model for Uncorrelated Noise Attenuation of Seismic Data. Geophysical Prospecting. 2026;74(9). doi:10.1111/1365-2478.70285
BibTeX
@article{Fu2026,
author = {Fu, Chao and Ma, Yu and Xu, Xiangtian},
title = {Pre‐Trained Foundation Model for Uncorrelated Noise Attenuation of Seismic Data},
journal = {Geophysical Prospecting},
year = {2026},
volume = {74},
number = {9},
doi = {10.1111/1365-2478.70285}
}
EUVIMED