Deep-learning inversion: A next-generation seismic velocity model building method

Yang, Fangshu; Ma, Jianwei

doi:10.1190/geo2018-0249.1

Cited by 388 publications

(153 citation statements)

References 56 publications

Supporting

Mentioning

123

Contrasting

Order By: Relevance

“…3. The calculated vertical and horizontal resolutions (Yilmaz, 2001) of the processed seismic section are approximately 1.5 and 17.3 m, respectively, when the central frequency is 250 Hz, sound speed is 1500 m s −1 , and reflector depth is 100 m. The internal waves in the research area propagate above a depth of 200 m, which is approximately 0.26 s in the seismic section. In addition, the physical properties of the research area were measured with oceanographic equipment, such as XCTD and XBT, during exploration.…”

Section: Sparker So Datamentioning

confidence: 94%

Random noise attenuation of sparker seismic oceanography data with machine learning

et al. 2020

View full text Add to dashboard Cite

Abstract. Seismic oceanography (SO) acquires water column reflections using controlled source seismology and provides high lateral resolution that enables the tracking of the thermohaline structure of the oceans. Most SO studies obtain data using air guns, which can produce acoustic energy below 100 Hz bandwidth, with vertical resolution of approximately 10 m or more. For higher-frequency bands, with vertical resolution ranging from several centimeters to several meters, a smaller, low-cost seismic exploration system may be used, such as a sparker source with central frequencies of 250 Hz or higher. However, the sparker source has a relatively low energy compared to air guns and consequently produces data with a lower signal-to-noise (S∕N) ratio. To attenuate the random noise and extract reliable signal from the low S∕N ratio of sparker SO data without distorting the true shape and amplitude of water column reflections, we applied machine learning. Specifically, we used a denoising convolutional neural network (DnCNN) that efficiently suppresses random noise in a natural image. One of the most important factors of machine learning is the generation of an appropriate training dataset. We generated two different training datasets using synthetic and field data. Models trained with the different training datasets were applied to the test data, and the denoised results were quantitatively compared. To demonstrate the technique, the trained models were applied to an SO sparker seismic dataset acquired in the Ulleung Basin, East Sea (Sea of Japan), and the denoised seismic sections were evaluated. The results show that machine learning can successfully attenuate the random noise in sparker water column seismic reflection data.

show abstract

Section: Sparker So Datamentioning

confidence: 94%

Random noise attenuation of sparker seismic oceanography data with machine learning

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Extra insights concerning the executions of these methods can be found in ( [12,13] and references in that). Figure 2-left shows the yield of the stationary wavelet transform (plotted on each other) for a few time windows X i removed from various seismograms [14][15][16][17][18]. The main two columns show the scaling capacity coefficients (W 1 ) for the L and P waves, separately.…”

Section: Classifiersmentioning

confidence: 99%

Influence of the Neutral Shadowing Force on the Pair Correlation Function of the Dusty Plasma Under Cryogenic Conditions

Aldakulov¹,

Temirbek²,

Muratov³

et al. 2021

SPM

View full text Add to dashboard Cite

In the course of recent years, progresses in sensor innovation has lead to increments in the interest for automated strategies for investigating seismological signals. Fundamental to the comprehension of the components creating seismic signals is the information on the phases of seismic waves. Having the option to indicate the kind of wave prompts better performing seismic forecasting frameworks. In this article, we propose another strategy for the characterization of seismic waves quantification from a three-channel seismograms. The seismograms are isolated into covering time windows, where each time-window is mapped to a lot of multi-scale three-dimensional unitary vectors that portray the direction of the seismic wave present in the window at a few physical scales. The issue of arranging seismic waves gets one of ordering focuses on a few two-dimensional unit circles. We take care of this issue by utilizing kernel based machine learning that are remarkably adjusted to the geometry of the circle. The grouping of the seismic wave depends on our capacity to gain proficiency with the limits between sets of focuses on the circles related with the various kinds of seismic waves. At each signal scale, we characterize a thought of vulnerability connected to the order that considers the geometry of the dissemination of tests on the circle. At long last, we join the grouping results acquired at each scale into a unique label.

show abstract

“…In geophysics, CNNs have initially been applied to aid structural interpretation of geophysical data such as seismic horizon and fault interpretation, and seismic texture identification (Xiong et al ., 2018; Waldeland et al ., 2018). Then, they have been extended to quantitatively solve geophysical problems such as velocity estimation (Araya‐Polo et al ., 2018), full‐waveform inversion (Lewis and Vigh, 2017; Richardson, 2018; Wu and McMechan, 2019; Yang and Ma, 2019), impedance inversion (Das et al ., 2019), electromagnetic inversion (Puzyrev, 2019), lithology prediction (Raeesi et al ., 2012; Hall, 2016), seismic deblending (Sun et al ., 2020), missing trace interpolation and noise attenuation (Liu et al ., 2018; Wang et al ., 2019; Mandelli et al ., 2019), multiple attenuation (Ma, 2018), pre‐stack seismic waveform classification and first‐break picking (Yuan et al ., 2018), automatic velocity analysis (Park and Sacchi, 2020) and reservoir characterization studies (Zhong et al ., 2019). In particular, Zhang et al .…”

Section: Introductionmentioning

confidence: 99%

Combining discrete cosine transform and convolutional neural networks to speed up the Hamiltonian Monte Carlo inversion of pre‐stack seismic data

Aleardi

2020

Geophysical Prospecting

View full text Add to dashboard Cite

Markov chain Monte Carlo algorithms are commonly employed for accurate uncertainty appraisals in non-linear inverse problems. The downside of these algorithms is the considerable number of samples needed to achieve reliable posterior estimations, especially in high-dimensional model spaces. To overcome this issue, the Hamiltonian Monte Carlo algorithm has recently been introduced to solve geophysical inversions. Different from classical Markov chain Monte Carlo algorithms, this approach exploits the derivative information of the target posterior probability density to guide the sampling of the model space. However, its main downside is the computational cost for the derivative computation (i.e. the computation of the Jacobian matrix around each sampled model). Possible strategies to mitigate this issue are the reduction of the dimensionality of the model space and/or the use of efficient methods to compute the gradient of the target density. Here we focus the attention to the estimation of elastic properties (P-, S-wave velocities and density) from pre-stack data through a non-linear amplitude versus angle inversion in which the Hamiltonian Monte Carlo algorithm is used to sample the posterior probability. To decrease the computational cost of the inversion procedure, we employ the discrete cosine transform to reparametrize the model space, and we train a convolutional neural network to predict the Jacobian matrix around each sampled model. The training data set for the network is also parametrized in the discrete cosine transform space, thus allowing for a reduction of the number of parameters to be optimized during the learning phase. Once trained the network can be used to compute the Jacobian matrix associated with each sampled model in real time. The outcomes of the proposed approach are compared and validated with the predictions of Hamiltonian Monte Carlo inversions in which a quite computationally expensive, but accurate finitedifference scheme is used to compute the Jacobian matrix and with those obtained by replacing the Jacobian with a matrix operator derived from a linear approximation of the Zoeppritz equations. Synthetic and field inversion experiments demonstrate that the proposed approach dramatically reduces the cost of the Hamiltonian Monte Carlo inversion while preserving an accurate and efficient sampling of the posterior probability.

show abstract

Deep-learning inversion: A next-generation seismic velocity model building method

Cited by 388 publications

References 56 publications

Random noise attenuation of sparker seismic oceanography data with machine learning

Random noise attenuation of sparker seismic oceanography data with machine learning

Influence of the Neutral Shadowing Force on the Pair Correlation Function of the Dusty Plasma Under Cryogenic Conditions

Combining discrete cosine transform and convolutional neural networks to speed up the Hamiltonian Monte Carlo inversion of pre‐stack seismic data

Contact Info

Product

Resources

About