Layer | Fill | Outline |
---|
Map layers
Theme | Visible | Selectable | Appearance | Zoom Range (now: 0) |
---|
Fill | Stroke |
---|---|
Collaborating Authors
Chen, Yangkang
ABSTRACT Localizing the microseismic event plays a key role in microseismic monitoring. However, microseismic data usually suffer from a low signal-to-noise ratio (S/N), which could affect the resolution of the microseismic source location. We have developed an unsupervised deep learning approach based on variational autoencoder (VAE) and squeeze-and-excitation (SE) networks for enhancing microseismic signals, as well as suppressing noise. First, the microseismic data are divided into several overlapped patches. Second, the VAE encodes the data, extracting the significant features related to the useful signals. Finally, the extracted latent features are decoded to uncover the useful signals and discard the others. The SE network is used to guide the VAE to preserve the useful information related to the clean signal by scaling the extracted features from the encoder part and concatenating them with the features of the decoder part. Our algorithm is evaluated using several synthetic and field examples. As a result, a robust denoising performance is shown despite the existence of a high level of random and coherent noise, for example, with an S/N as low as โ32.45ย dB. Then, the denoised signal can be used as input data to image the source location using a reverse time migration method, leading to better location accuracy. Our algorithm performs the best when compared to benchmark methods such as f-x deconvolution and the damped multichannel singular spectrum analysis methods.
Local seismic attributes play an important role in seismic processing and interpretation. In this abstract, we present an efficient method for estimating local seismic attributes, including local frequency and local spectrum, using streaming computation. In our proposed approach, the local attributes can be computed by updating the previously calculated attribute value using a new data point at a time. We apply the proposed method to both synthetic and field data to demonstrate its efficiency and effectiveness to accurately characterize nonstationary seismic signals.
- Geophysics > Seismic Surveying > Seismic Processing (1.00)
- Geophysics > Seismic Surveying > Seismic Interpretation (1.00)
Dictionary learning with convolutional structure for seismic data denoising and interpolation
Almadani, Murad (King Fahd University of Petroleum and Minerals) | Waheed, Umair bin (King Fahd University of Petroleum and Minerals) | Masood, Mudassir (King Fahd University of Petroleum and Minerals) | Chen, Yangkang (Zhejiang University)
ABSTRACT Seismic data inevitably suffer from random noise and missing traces in field acquisition. This limits the use of seismic data for subsequent imaging or inversion applications. Recently, dictionary learning has gained remarkable success in seismic data denoising and interpolation. Variants of the patch-based learning technique, such as the K-singular value decomposition (K-SVD) algorithm, have been shown to improve denoising and interpolation performance compared with the analytic transform-based methods. However, patch-based learning algorithms work on overlapping patches of data and do not take the full data into account during reconstruction. In contrast, the data patches (convolutional sparse coding [CSC]) model treats signals globally and, therefore, has shown superior performance over patch-based methods in several image processing applications. As a consequence, we test use of the CSC model for seismic data denoising and interpolation. In particular, we use the local block coordinate descent (LoBCoD) algorithm to reconstruct missing traces and clean seismic data from noisy input. The denoising and interpolation performance of the LoBCoD algorithm has been compared with that of K-SVD and orthogonal matching pursuit (OMP) algorithms using synthetic and field data examples. We have used three quality measures to test the denoising accuracy: the peak signal-to-noise ratio (PS/N), the relative L2-norm of the error (RLNE), and the structural similarity index (SSIM). We find that LoBCoD performs better than K-SVD and OMP for all test cases in improving PS/N and SSIM and in reducing RLNE. These observations suggest a huge potential of the CSC model in seismic data denoising and interpolation applications.
ABSTRACT The local signal-and-noise orthogonalization method has been widely used in the seismic processing and imaging community. This method uses a fixed triangle smoother for regularizing the local orthogonalization weight, which is based on the assumption that the energy is homogeneously distributed across the whole seismic profile. The fixed triangle smoother limits the performance of the local orthogonalization method in processing complicated seismic data sets. We have developed a new local orthogonalization method that uses a variable triangle smoother. The nonstationary smoothing radius is obtained by solving an optimization problem in which the low-pass-filtered seismic data are matched by the smoothed data in terms of the local frequency attribute. The new local orthogonalization method with the nonstationary model smoothness constraint is called the nonstationary local orthogonalization method. We have proven the successful performance of the new method using several synthetic and field data evaluations.
5D dealiased seismic data interpolation using nonstationary prediction-error filter
Chen, Yangkang (The University of Texas at Austin, John A. and Katherine G. Jackson School of Geosciences) | Fomel, Sergey (The University of Texas at Austin, John A. and Katherine G. Jackson School of Geosciences) | Wang, Hang (Zhejiang University) | Zu, Shaohuan (Chengdu University of Technology)
ABSTRACT The prediction-error filter (PEF) assumes that seismic data can be destructed to zero by applying a convolutional operation between the target data and the prediction filter in either the time-space or frequency-space domain. We have extended the commonly known PEF in 2D or 3D problems to its 5D version. To handle the nonstationary property of the seismic data, we formulate the PEF in a nonstationary way, which is called the nonstationary prediction-error filter (NPEF). In NPEF, the coefficients of a fixed-size PEF vary across the whole seismic data. In NPEF, we aim at solving a highly ill-posed inverse problem via the computationally efficient iterative shaping regularization. NPEF can be used to denoise multidimensional seismic data and, more importantly, to restore the highly incomplete aliased 5D seismic data. We compare our NPEF method with the state-of-the-art rank-reduction method for the 5D seismic data interpolation in cases of irregularly and regularly missing traces via several synthetic and real seismic data. The results show that although our NPEF method is less effective than the rank-reduction method in interpolating irregularly missing traces especially in the case of a low signal-to-noise ratio, it outperforms the rank-reduction method in interpolating an aliased 5D data set with regularly missing traces.
- Research Report > New Finding (0.34)
- Research Report > Experimental Study (0.34)
ABSTRACT Noise and missing traces usually influence the quality of multidimensional seismic data. Therefore, it is necessary to estimate the useful signal from its noisy observation. The damped rank-reduction (DRR) method has emerged as an effective method to reconstruct the useful signal matrix from noisy and incomplete observations. However, the higher the noise level and the larger the ratio of missing traces, the weaker the DRR operator becomes. Consequently, the estimated low-rank (LR) signal matrix includes a significant amount of residual noise that influences the following processing steps. Therefore, we focus on the problem of estimating an LR signal matrix from its noisy observation. To elaborate on the novel algorithm, we formulate an improved proximity function by mixing the moving-average filter and the arctangent penalty function. First, we apply the proximity function to the level-4 block Hankel matrix before singular-value decomposition (SVD) and, then, to singular values, during the damped truncated SVD process. The combination of the novel proximity function and the DRR framework leads to an optimization problem, which results in better recovery performance. Our algorithm aims at producing an enhanced rank-reduction operator to estimate the useful signal matrix with higher quality. Experiments are conducted on synthetic and real 5D seismic data to compare the effectiveness of our approach to the DRR approach. Our approach obtains better performance because the estimated LR signal matrix is cleaner and contains fewer artifacts compared to that reconstructed by the DRR algorithm.
- Geophysics > Seismic Surveying > Seismic Processing (1.00)
- Geophysics > Seismic Surveying > Surface Seismic Acquisition (0.68)
ABSTRACT Surface-related multiple elimination (SRME) has proven to be a robust surface multiple and primary estimation tool for decades. However, surface-related multiple leakage is still commonly observed in SRME-processed results due to imperfect multiple predictions. Usually, adaptive subtraction cannot fully correct for these effects without primary damage. Local primary-and-multiple orthogonalization (LPMO) has recently been proposed to partially mitigate surface multiple leakage, by multiplication of the estimated primaries with a weight function that scales down residual multiples while preserving primaries. The weight function is determined by shaping regularization followed by thresholding and median filtering. Although effective leakage extraction can be achieved, LPMO has a large computational cost due to many conjugate-gradient iterations within the shaping regularization-based inversion framework. Using information on the typical local coherency length of primaries and multiples, a spatially constrained scaled point-by-point division can be used to avoid the iterative inversion within the LPMO method. Based on this, we have adopted a fast LPMO (FLPMO) for surface-related multiple leakage estimation and extraction. Applications on two different field data sets demonstrate the very similar surface multiple leakage extraction performance for LPMO and FLPMO, while showing that the scaled point-by-point division in FLPMO is approximately 40 times faster on real data sets than the shaping regularization-based inversion in LPMO.
- Europe (1.00)
- North America > United States > Texas (0.28)
- Europe > United Kingdom > North Sea > Central North Sea > Central Graben > Block 22/7 > Nelson Field > Forties Formation (0.99)
- Europe > United Kingdom > North Sea > Central North Sea > Central Graben > Block 22/6a > Nelson Field > Forties Formation (0.99)
- Europe > United Kingdom > North Sea > Central North Sea > Central Graben > Block 22/12a > Nelson Field > Forties Formation (0.99)
- (3 more...)
Training deep networks with only synthetic data: Deep-learning-based near-offset reconstruction for (closed-loop) surface-related multiple estimation on shallow-water field data
Qu, Shan (Delft University of Technology) | Verschuur, Eric (Delft University of Technology) | Zhang, Dong (Delft University of Technology) | Chen, Yangkang (Zhejiang University)
ABSTRACT Accurate removal of surface-related multiples remains a challenge in shallow-water cases. One reason is that the success of surface-related multiple estimation (SRME)-related algorithms is sensitive to the quality of the near-offset reconstruction. When it comes to a larger missing gap and a shallower water bottom, the state-of-the-art near-offset gap construction method โ the parabolic Radon transform โ fails to provide reliable recovery of the shallow reflections due to the limited information from the data and highly curved events at near offsets with strong lateral amplitude variations. Therefore, we have developed a novel workflow that first deploys a deep-learning-based reconstruction of the shallow reflections and then uses the reconstructed data as the input for the subsequent surface multiple removal. In particular, we use a convolutional neural network architecture โ U-net that was developed from convolutional autoencoders with extra direct skip connections between different levels of encoders and the corresponding decoders. Instead of using field data directly in network training, the training set is carefully synthesized based on the prior water-layer information of the field data; thus, a fully sampled field data set, which is difficult to obtain, is not needed for training in our workflow. An inversion-based approach โ closed-loop SRME โ is used for the surface multiple removal, in which the primaries are directly estimated via full-waveform inversion in a data-driven manner. Finally, the effectiveness of our workflow is determined based on 2D North Sea field data in a shallow-water scenario (92.5ย m water depth) with a relatively large minimum offset (150ย m).
- Geophysics > Seismic Surveying > Seismic Processing (1.00)
- Geophysics > Seismic Surveying > Seismic Modeling > Velocity Modeling (0.34)
Nonstationary predictive filtering for seismic random noise suppression โ A tutorial
Wang, Hang (Zhejiang University) | Chen, Wei (Yangtze University) | Huang, Weilin (China University of Petroleum) | Zu, Shaohuan (Chengdu University of Technology) | Liu, Xingye (Xiโan University of Science and Technology) | Yang, Liuqing (China University of Petroleum) | Chen, Yangkang (Zhejiang University)
ABSTRACT Predictive filtering (PF) in the frequency domain is one of the most widely used denoising algorithms in seismic data processing. PF is based on the assumption of linear or planar events in the time-space domain. In traditional PF methods, a predictive filter is fixed across the spatial dimension, which cannot deal with spatial variations in seismic data well. To handle the curved events, the predictive filter is either applied in local windows or extended into a nonstationary version. The regularized nonstationary autoregression (RNAR) method can be treated as a nonstationary extension of traditional PF, in which the predictive filter coefficients are variable in different spatial locations. This highly underdetermined inverse problem is solved by shaping regularization with a smoothness constraint in space. We further extend the RNAR method to the more general case, in which we can apply more constraints to the filter coefficients according to the features of seismic data. First, apart from the smoothness in space, we also apply a smoothing constraint in frequency, considering the coherency of the coefficients in the frequency dimension. Second, we apply a frequency-dependent smoothing radius in the spatial dimension to better take advantage of the nonstationarity of seismic data in the frequency axis and to better deal with noise. The effectiveness of our method is validated using several synthetic and field data examples.
ABSTRACT Time-frequency analysis is a fundamental approach to many seismic problems. Time-frequency decomposition transforms input seismic data from the time domain to the time-frequency domain, offering a new dimension to probe the hidden information inside the data. Considering the nonstationary nature of seismic data, time-frequency spectra can be obtained by applying a local time-frequency transform (LTFT) method that matches the input data by fitting the Fourier basis with nonstationary Fourier coefficients in the shaping regularization framework. The key part of LTFT is the temporal smoother with a fixed smoothing radius that guarantees the stability of the nonstationary least-squares fitting. We have developed a new LTFT method to handle the nonstationarity in all time, frequency, and space (x and y) directions of the input seismic data by extending fixed-radius temporal smoothing to nonstationary smoothing with a variable radius in all physical dimensions. The resulting time-frequency transform is referred to as the nonstationary LTFT method, which could significantly increase the resolution and antinoise ability of time-frequency transformation. There are two meanings of nonstationarity, i.e., coping with the nonstationarity in the data by LTFT and dealing with the nonstationarity in the model by nonstationary smoothing. We evaluate the performance of our nonstationary LTFT method in several standard seismic applications via synthetic and field data sets, e.g., arrival picking, quality factor estimation, low-frequency shadow detection, channel detection, and multicomponent data registration, and we benchmark the results with the traditional stationary LTFT method.