Deep-learning based ocean bottom seismic wavefield recovery

Siahkoohi, Ali; Kumar, Rajiv; Herrmann, Felix J.

doi:10.1190/segam2019-3216632.1

Cited by 16 publications

(6 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Including the exact source effects, azimuthal differences and rough sea surface effects in the training data could be challenging and further research is required. In addition, comparing our proposed method with existing deep learning interpolation schemes (Wang et al, 2020;Siahkoohi et al, 2019a;Garg et al, 2019) followed by, e.g., a sparse source deghosting method could also provide more insight in its capabilities.…”

Section: Discussionmentioning

confidence: 96%

Source deghosting of coarsely sampled common-receiver data using a convolutional neural network

Vrolijk

Blacquière

2021

GEOPHYSICS

View full text Add to dashboard Cite

It is well known that source deghosting can best be applied to common-receiver gathers, while receiver deghosting can best be applied to common-shot records. The source-ghost wavefield observed in the common-shot domain contains the imprint of the subsurface, which complicates source deghosting in common-shot domain, in particular when the subsurface is complex. Unfortunately, the alternative, i.e., the common-receiver domain, is often coarsely sampled, which complicates source deghosting in this domain as well. To solve the latter issue, we propose to train a convolutional neural network to apply source deghosting in this domain. We subsample all shot records with and without the receiver ghost wavefield to obtain the training data. Due to reciprocity this training data is a representative data set for source deghosting in the coarse common-receiver domain. We validate the machine-learning approach on simulated data and on field data. The machine learning approach gives a significant uplift to the simulated data compared to conventional source deghosting. The field-data results confirm that the proposed machine-learning approach is able to remove the source-ghost wavefield from the coarsely-sampled common-receiver gathers.

show abstract

Section: Discussionmentioning

confidence: 96%

Source deghosting of coarsely sampled common-receiver data using a convolutional neural network

Vrolijk

Blacquière

2021

GEOPHYSICS

View full text Add to dashboard Cite

show abstract

“…To address the seismic wavefield's 3D nature, Siahkoohi et al (2019b) trained a Generative Adversarial Networks (GAN) using pairs of fully-sampled and under-sampled single-receiver frequency slices. They showed promising results for synthetic ocean bottom node data.…”

Section: Shot Point Interpolation Using Deep Neural Networkmentioning

confidence: 99%

Noise types and their attenuation in towed marine seismic: A tutorial

Hlebnikov

Elboth²,

Vinje³

et al. 2021

GEOPHYSICS

View full text Add to dashboard Cite

The presence of noise in towed marine seismic data is a long-standing problem. The various types of noise present in marine seismic records are never truly random. Instead, seismic noise is more complex and often challenging to attenuate in seismic data processing. Therefore, we examine a wide range of real data examples contaminated by different types of noise including swell noise, seismic interference noise, strumming noise, passing vessel noise, vertical particle velocity noise, streamer hit and fishing gear noise, snapping shrimp noise, spike-like noise, cross-feed noise and streamer mounted devices noise. The noise examples investigated focus only on data acquired with analogue group-forming. Each noise type is classified based on its origin, coherency and frequency content. We then demonstrate how the noise component can be effectively attenuated through industry standard seismic processing techniques. In this tutorial, we avoid presenting the finest details of either the physics of the different types of noise themselves or the noise attenuation algorithms applied. Rather, we focus on presenting the noise problems themselves and show how well the community is able to address such noise. Our aim is that based on the provided insights, the geophysical community will be able to gain an appreciation of some of the most common types of noise encountered in marine towed seismic, in the hope to inspire more researchers to focus their attention on noise problems with greater potential industry impact.

show abstract

“…For example, Liu et al (2018) proposed the use of partial convolution methods to improve the blur problem of reconstructed images. Siahkoohi et al (2019) accomplished the accurate reconstruction of the common shot records by the CNN, which is trained by the common receiver records in the FK domain based on the reciprocity theorem. Chen and Wang, (2021) proposed a method to enrich the training set by sampling at different scales and image flipping to improve the generalization ability of CNN in seismic data reconstruction.…”

Section: Introductionmentioning

confidence: 99%

Simultaneous reconstruction and denoising for DAS-VSP seismic data by RRU-net

Tang

Cheng

et al. 2023

Front. Earth Sci.

View full text Add to dashboard Cite

Distributed acoustic sensing in vertical seismic profile (DAS-VSP) acquisition plays an important role in reservoir monitoring. But the field data can be noisy and associated with missing traces which affects the seismic imaging and geological interpretation. Therefore, the DAS-VSP seismic data reconstruction with a high signal-to-noise ratio (SNR) is worth studying. There are no exact relationships between signals and noise in the t-x domain DAS-VSP seismic data, which means that reconstructing signals and suppressing noise simultaneously by the deep neural network is difficult. We develop a novel algorithm based on U-net in combination with the Hankel matrix as input/output, rather than t-x domain seismic data. The frequency domain Hankel matrix of the seismic data is proposed to facilitate the reconstruction and denoising of DAS-VSP seismic data as a rank reduction problem of the high-rank matrix. The Hankel matrices of incomplete data with noise are high-rank ones while those of complete data without noise are low-rank ones, which is beneficial to the network learning. In our proposed rank reduction U-net (RRU-net), two-channel input/output layers are designed for the real and the imaginary parts of the Hankel matrix in the frequency domain. Thus, reconstructed data with high precision and high SNR could be obtained using a trained RRU-net. Meanwhile, we tested our RRU-net algorithm on two synthetic data and one field data, and the results show the effectiveness and the feasibility of the method. Our algorithm performs better than both the U-net-based method that uses t−x domain data as input/output and the rank reduction approach.

show abstract

Deep-learning based ocean bottom seismic wavefield recovery

Cited by 16 publications

References 19 publications

Source deghosting of coarsely sampled common-receiver data using a convolutional neural network

Source deghosting of coarsely sampled common-receiver data using a convolutional neural network

Noise types and their attenuation in towed marine seismic: A tutorial

Simultaneous reconstruction and denoising for DAS-VSP seismic data by RRU-net

Contact Info

Product

Resources

About