Full text loading...
-
Unsupervised Learning for Ground Roll Attenuation with Frequency Band Similarity
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 86th EAGE Annual Conference & Exhibition, Jun 2025, Volume 2025, p.1 - 5
Abstract
Summary
Ground roll, characterized by low frequency, low velocity, and strong energy, can obscure the effective signal and degrade the signal-to-noise ratio. Traditional methods, such as frequency-wavenumber (F-K) filtering, often leave residual ground roll. To achieve complete attenuation of ground roll and prevent leakage of effective signals, we propose an unsupervised learning method based on frequency band similarity, combined with F-K filtering pre-processing (UL-FBS-FK). First, F-K filtering is used to suppress ground roll and obtain the preliminary suppression result (PSR). The mid-frequency band (FB) and high-FB PSRs predominantly contain effective signals. Despite differences in amplitudes, the mid-and the low-to-mid-FB effective signals exhibit morphological and kinematic similarities. Our unsupervised learning (UL) method does not require clean labelled data or traditional methods for training labels. We apply UL to map mid-FB effective signals to low-to-mid-FB effective signals, further suppressing residual ground roll. Field data results show that our approach significantly outperforms the traditional F-K filtering method in suppressing ground roll.
© EAGE Publications BV