Extrapolating low-frequency prestack land data with deep learning

Ovcharenko, Oleg; Kazei, Vladimir; Plotnitskiy, Pavel; Peter, Daniel; Silvestrov, Ilya; Bakulin, Andrey; Alkhalifah, Tariq

doi:10.1190/segam2020-3427522.1

Cited by 17 publications

(3 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The approach is suitable for elastic waveform inversion in marine survey layout. Fang et al [42] and Ovcharenko et al [43] extrapolated low frequencies by training convolutional networks on AGC-balanced patches of time-domain seismic data in marine and land setups, respectively. Aharchaou and Baumstein [44] avoided using synthetic data and trained a UNet neural network to translate knowledge of low-frequency data from OBN surveys to band-limited shallow streamer data.…”

Section: B Reconstruction Of Missing Low-frequency Datamentioning

confidence: 99%

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Ovcharenko

Kazei

Alkhalifah

et al. 2022

IEEE Trans. Geosci. Remote Sensing

Self Cite

View full text Add to dashboard Cite

Low-frequency signal content in seismic data as well as a realistic initial model are key ingredients for robust and efficient full-waveform inversions. However, acquiring low-frequency data is challenging in practice for active seismic surveys. Data-driven solutions show promise to extrapolate low-frequency data given a high-frequency counterpart. While being established for synthetic acoustic examples, the application of bandwidth extrapolation to field datasets remains non-trivial. Rather than aiming to reach superior accuracy in bandwidth extrapolation, we propose to jointly reconstruct low-frequency data and a smooth background subsurface model within a multi-task deep learning framework. We automatically balance data, model and trace-wise correlation loss terms in the objective functional and show that this approach improves the extrapolation capability of the network. We also design a pipeline for generating synthetic data suitable for field data applications. Finally, we apply the same trained network to synthetic and real marine streamer datasets and run an elastic full-waveform inversion from the extrapolated dataset.

show abstract

Section: B Reconstruction Of Missing Low-frequency Datamentioning

confidence: 99%

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Ovcharenko

Kazei

Alkhalifah

et al. 2022

IEEE Trans. Geosci. Remote Sensing

Self Cite

View full text Add to dashboard Cite

show abstract

“…To mitigate potential bias toward the more energetic aspects of seismic records, amplitude balancing techniques are introduced, including automatic gain control [21]. Further, a frequency band expansion method rooted in the time-space domain was explored [22], with an emphasis on utilizing low-frequency remote offset seismic records for training neural networks and predicting the low-frequency components. However, this approach can amplify both weak energy seismic signals and noise, adversely affecting prediction outcomes.…”

Section: Introductionmentioning

confidence: 99%

A Multi-Task Learning Framework of Stable Q-Compensated Reverse Time Migration Based on Fractional Viscoacoustic Wave Equation

Xue,

Ma,

Wang

et al. 2023

Fractal Fract

View full text Add to dashboard Cite

Q-compensated reverse time migration (Q-RTM) is a crucial technique in seismic imaging. However, stability is a prominent concern due to the exponential increase in high-frequency ambient noise during seismic wavefield propagation. The two primary strategies for mitigating instability in Q-RTM are regularization and low-pass filtering. Q-RTM instability can be addressed through regularization. However, determining the appropriate regularization parameters is often an experimental process, leading to challenges in accurately recovering the wavefield. Another approach to control instability is low-pass filtering. Nevertheless, selecting the cutoff frequency for different Q values is a complex task. In situations with low signal-to-noise ratios (SNRs) in seismic data, using low-pass filtering can make Q-RTM highly unstable. The need for a small cutoff frequency for stability can result in a significant loss of high-frequency signals. In this study, we propose a multi-task learning (MTL) framework that leverages data-driven concepts to address the issue of amplitude attenuation in seismic records, particularly when dealing with instability during the Q-RTM (reverse time migration with Q-attenuation) process. Our innovative framework is executed using a convolutional neural network. This network has the capability to both predict and compensate for the missing high-frequency components caused by Q-effects while simultaneously reconstructing the low-frequency information present in seismograms. This approach helps mitigate overwhelming instability phenomena and enhances the overall generalization capacity of the model. Numerical examples demonstrate that our Q-RTM results closely align with the reference images, indicating the effectiveness of our proposed MTL frequency-extension method. This method effectively compensates for the attenuation of high-frequency signals and mitigates the instability issues associated with the traditional Q-RTM process.

show abstract

“…Aharchaou et al (2020) showed a field data application on ocean-bottom node data. Elastic land data was extrapolated in synthetic setup in (Ovcharenko et al, 2020;Sun and Demanet, 2020). Low-wavenumber velocity model was recovered from early FWI gradients and directly from the data in Plotnitskii et al (2020) and Kazei et al (2019b), respectively.…”

mentioning

confidence: 99%

Transferring Elastic Low Frequency Extrapolation from Synthetic to Field Data

Ovcharenko¹,

Kazei

Peter³

et al. 2021

82nd EAGE Annual Conference &Amp; Exhibition

View full text Add to dashboard Cite

Training deep learning models on synthetic data is a common practice in geophysics. However, knowledge transfer from synthetic to field applications is often a bottleneck. Here, we describe the workflow for generation of realistic synthetic dataset of elastic waveforms, sufficient for low-frequency extrapolation in marine streamer setup. Namely, we first extract the source signature, the noise imprint and a 1D velocity model from real marine data. Then, we use those to generate pseudo-random initializations of elastic subsurface models and simulate elastic wavefield data. After that, we enrich the simulated data with realistic noise and use it to train a deep neural network. Finally, we demonstrate the results of low-frequency extrapolation on field streamer data, given that the model was trained exclusively on a synthetic dataset.

show abstract

Extrapolating low-frequency prestack land data with deep learning

Cited by 17 publications

References 31 publications

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

Multi-Task Learning for Low-Frequency Extrapolation and Elastic Model Building From Seismic Data

A Multi-Task Learning Framework of Stable Q-Compensated Reverse Time Migration Based on Fractional Viscoacoustic Wave Equation

Transferring Elastic Low Frequency Extrapolation from Synthetic to Field Data

Contact Info

Product

Resources

About