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

Geophysics ◽  
2021 ◽  
pp. 1-32
Author(s):  
Shan Qu ◽  
Eric Verschuur ◽  
Dong Zhang ◽  
Yangkang Chen
Keyword(s):  
Deep Learning ◽  
Shallow Water ◽  
Network Architecture ◽  
Field Data ◽  
Closed Loop ◽  
Waveform Inversion ◽  
Synthetic Data ◽  
Water Layer ◽  
Limited Information ◽  
Loop Surface

Accurate removal of surface-related multiples remains a challenge in shallow-water cases. One reason is that the success of the 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 — parabolic Radon transform (PRT) — fails to provide a 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 propose a novel workflow, which first deploys a deep-learning(DL)-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 --- U-net, which 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 dataset, which is hard to obtain, is not needed for training in the proposed workflow. An inversion-based approach — closed-loop surface-related multiple estimation (CL-SRME) -- 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 the proposed workflow is demonstrated based on a 2D North Sea field data in a shallow-water scenario (92.5 m water depth) with a relatively large minimum offset (150 m).

Closed-loop surface-related multiple estimation with full-wavefield migration-reconstructed near-offsets for shallow water

Geophysics ◽  
2021 ◽  
pp. 1-72
Author(s):  
Dong Zhang ◽  
Dirk Jacob (Eric) Verschuur
Keyword(s):  
Shallow Water ◽  
Closed Loop ◽  
Good Description ◽  
Water Environment ◽  
Earth Model ◽  
Multiple Reflections ◽  
Strong Constraint ◽  
Integrated Framework ◽  
Shallow Water Environment ◽  
Loop Surface

Reliably separating primary and multiple reflections in a shallow water environment (i.e., 50 m to 200 m water depth) still remains a challenge. The success of previously published closed-loop surface-related multiple estimation (CL-SRME) depends heavily on the data coverage, i.e., the near-offset reconstruction. Therefore, we propose the integrated framework of CL-SRME and full-wavefield migration (FWM). Multiples recorded in the data are capable of helping infill the acquisition imprint of the FWM image. With this image as a strong constraint, we are able to reconstruct the data at near-offsets, which is essential for better primary and multiple estimation during CL-SRME. FWM applied in a non-linear way can avoid the negative influences from the missing data, and at the same time bring in more physics between primaries and multiples. The FWM image of the top part of the subsurface is also used to back-project the information from multiples to primaries with the physical constraint of all this information belongs to the same earth model, provided that a good description of the source wavefield and a reasonable velocity model are available. The proposed integrated framework first reconstructs near-offsets via the closed-loop imaging process of FWM and then feeds the complete reconstructed data to CL-SRME for better primary and multiple estimation. A good performance is demonstrated on both 2D synthetic and field data examples in a challenging shallow water environment.

scholarly journals Attenuation of Seismic Multiples in Very Shallow Water: An Application in Archaeological Prospection Using Data Driven Approaches

2021 ◽  
Vol 13 (10) ◽  
pp. 1871
Author(s):  
Michaela Schwardt ◽  
Dennis Wilken ◽  
Wolfgang Rabbel
Keyword(s):  
Shallow Water ◽  
Field Data ◽  
Water Layer ◽  
Primary Energy ◽  
Time Shift ◽  
Multiple Reflections ◽  
Traditional Methods ◽  
Near Surface ◽  
Archaeological Prospection ◽  
Model Driven

Water-layer multiples pose a major problem in shallow water seismic investigations as they interfere with primaries reflected from layer boundaries or archaeology buried only a few meters below the water bottom. In the present study we evaluate two model-driven approaches (“Prediction and Subtraction” and “RTM-Deco”) to attenuate water-layer multiple reflections in very shallow water using synthetic and field data. The tests comprise both multi- and constant-offset data. We compare the multiple removal efficiency of the evaluated methods with two traditional methods (Predictive Deconvolution and SRME). Both model-driven approaches yield satisfactory results concerning the enhancement of primary energy and the attenuation of multiple energy. For the synthetic test cases, the multiple energy is reduced by at least 80% for the Prediction and Subtraction approach, and by more than 60% for the RTM-Deco approach. The application to two field data sets shows a significant amplification of primaries formerly hidden by the first water-layer multiple, with a reduction of multiple energy of up to 50%. The waveforms obtained from FD modeling match the true waveforms of the field data well and small deviations in time and amplitude can be removed by a time shift of the traces as well as an amplitude adaption to the field data. The field data examples should be emphasized, where the tested Prediction and Subtraction approach works significantly better than the traditional methods: the multiples are effectively predicted and attenuated while primary signals are highlighted. In conclusion, this shows that this method is particularly suitable in shallow water applications. Both evaluated multiple attenuation approaches could be successfully transferred to two other 3D systems used in shallow water near surface investigations. Especially the Prediction and Subtraction approach is able to enhance the primaries for both tested 3D systems with the multiple energy being reduced by more than 50%.

scholarly journals Connected-UNets: a deep learning architecture for breast mass segmentation

2021 ◽  
Vol 7 (1) ◽  
Author(s):  
Asma Baccouche ◽  
Begonya Garcia-Zapirain ◽  
Cristian Castillo Olea ◽  
Adel S. Elmaghraby
Keyword(s):  
Deep Learning ◽  
Network Architecture ◽  
Breast Imaging ◽  
Rapid Development ◽  
Synthetic Data ◽  
Screening Mammography ◽  
Medical Image Segmentation ◽  
Breast Mass ◽  
Generative Adversarial Network ◽  
Mass Segmentation

AbstractBreast cancer analysis implies that radiologists inspect mammograms to detect suspicious breast lesions and identify mass tumors. Artificial intelligence techniques offer automatic systems for breast mass segmentation to assist radiologists in their diagnosis. With the rapid development of deep learning and its application to medical imaging challenges, UNet and its variations is one of the state-of-the-art models for medical image segmentation that showed promising performance on mammography. In this paper, we propose an architecture, called Connected-UNets, which connects two UNets using additional modified skip connections. We integrate Atrous Spatial Pyramid Pooling (ASPP) in the two standard UNets to emphasize the contextual information within the encoder–decoder network architecture. We also apply the proposed architecture on the Attention UNet (AUNet) and the Residual UNet (ResUNet). We evaluated the proposed architectures on two publically available datasets, the Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM) and INbreast, and additionally on a private dataset. Experiments were also conducted using additional synthetic data using the cycle-consistent Generative Adversarial Network (CycleGAN) model between two unpaired datasets to augment and enhance the images. Qualitative and quantitative results show that the proposed architecture can achieve better automatic mass segmentation with a high Dice score of 89.52%, 95.28%, and 95.88% and Intersection over Union (IoU) score of 80.02%, 91.03%, and 92.27%, respectively, on CBIS-DDSM, INbreast, and the private dataset.

scholarly journals Waveform inversion using a logarithmic wavefield

Geophysics ◽  
2006 ◽  
Vol 71 (3) ◽  
pp. R31-R42 ◽  
Author(s):  
Changsoo Shin ◽  
Dong-Joo Min
Keyword(s):  
Objective Function ◽  
Field Data ◽  
Velocity Structure ◽  
Waveform Inversion ◽  
Synthetic Data ◽  
Velocity Model ◽  
Matrix Formalism ◽  
Inversion Algorithm ◽  
Data Set ◽  
Short Period

Although waveform inversion has been studied extensively since its beginning [Formula: see text] ago, applications to seismic field data have been limited, and most of those applications have been for global-seismology- or engineering-seismology-scale problems, not for exploration-scale data. As an alternative to classical waveform inversion, we propose the use of a new, objective function constructed by taking the logarithm of wavefields, allowing consideration of three types of objective function, namely, amplitude only, phase only, or both. In our wave form inversion, we estimate the source signature as well as the velocity structure by including functions of amplitudes and phases of the source signature in the objective function. We compute the steepest-descent directions by using a matrix formalism derived from a frequency-domain, finite-element/finite-difference modeling technique. Our numerical algorithms are similar to those of reverse-time migration and waveform inversion based on the adjoint state of the wave equation. In order to demonstrate the practical applicability of our algorithm, we use a synthetic data set from the Marmousi model and seismic data collected from the Korean continental shelf. For noise-free synthetic data, the velocity structure produced by our inversion algorithm is closer to the true velocity structure than that obtained with conventional waveform inversion. When random noise is added, the inverted velocity model is also close to the true Marmousi model, but when frequencies below [Formula: see text] are removed from the data, the velocity structure is not as good as those for the noise-free and noisy data. For field data, we compare the time-domain synthetic seismograms generated for the velocity model inverted by our algorithm with real seismograms and find that the results show that our inversion algorithm reveals short-period features of the subsurface. Although we use wrapped phases in our examples, we still obtain reasonable results. We expect that if we were to use correctly unwrapped phases in the inversion algorithm, we would obtain better results.

scholarly journals Near-surface diffractor detection at archaeological sites based on an interferometric workflow

Geophysics ◽  
2021 ◽  
pp. 1-75
Author(s):  
Jianhuan Liu ◽  
Deyan Draganov ◽  
Ranajit Ghose ◽  
Quentin Bourgeois
Keyword(s):  
Field Data ◽  
Waveform Inversion ◽  
Synthetic Data ◽  
Spatial Summation ◽  
Archaeological Site ◽  
Archaeological Sites ◽  
Magnetic Survey ◽  
Seismic Reflection Data ◽  
Near Surface ◽  
Reflection Data

Detecting small-size objects is a primary challenge at archaeological sites due to the high degree of heterogeneity present in the near surface. Although high-resolution reflection seismic imaging often delivers the target resolution of the subsurface in different near-surface settings, the standard processing for obtaining an image of the subsurface is not suitable to map local diffractors. This happens because shallow seismic-reflection data are often dominated by strong surface waves which might cover weaker diffractions, and because traditional common-midpoint moveout corrections are only optimal for reflection events. Here, we propose an approach for imaging subsurface objects using masked diffractions. These masked diffractions are firstly revealed by a combination of seismic interferometry and nonstationary adaptive subtraction, and then further enhanced through crosscoherence-based super-virtual interferometry. A diffraction image is then computed by a spatial summation of the revealed diffractions. We use phase-weighted stack to enhance the coherent summation of weak diffraction signals. Using synthetic data, we show that our scheme is robust in locating diffractors from data dominated by strong Love waves. We test our method on field data acquired at an archaeological site. The resulting distribution of shallow diffractors agrees with the location of anomalous objects identified in the Vs model obtained by elastic SH/Love full-waveform inversion using the same field data. The anomalous objects correspond to the position of a suspected burial, also detected in an independent magnetic survey and corings.

Applications of multiscale waveform inversion to marine data using a flooding technique and dynamic early-arrival windows

Geophysics ◽  
2010 ◽  
Vol 75 (6) ◽  
pp. R129-R136 ◽  
Author(s):  
Chaiwoot Boonyasiriwat ◽  
Gerard T. Schuster ◽  
Paul Valasek ◽  
Weiping Cao
Keyword(s):  
Hybrid Method ◽  
Field Data ◽  
Waveform Inversion ◽  
Synthetic Data ◽  
Convergence Problem ◽  
Early Arrival ◽  
Data Application ◽  
Marine Data ◽  
Waveform Tomography ◽  
Subsalt Imaging

A recently developed time-domain multiscale waveform tomography (MWT) method is applied to synthetic and field marine data. Although the MWT method was already applied to synthetic data, the synthetic data application leads to a development of a hybrid method between waveform tomography and the salt flooding technique commonly use in subsalt imaging. This hybrid method can overcome a convergence problem encountered by inversion with a traveltime velocity tomogram and successfully provides an accurate and highly resolved velocity tomogram for the 2D SEG/EAGE salt model. In the application of MWT to the field data, the inversion process is carried out using a multiscale method with a dynamic early-arrival muting window to mitigate the local minima problem of waveform tomography and elastic effects. With the modified MWT method, reasonably accurate results as verified by comparison of migration images and common image gathers were obtained. The hybrid method with the salt flooding technique is not used in this field data example because there is no salt in the subsurface according to our interpretation. However, we believe it is applicable to field data applications.

Fast least-squares migration with a deblurring filter

Geophysics ◽  
2009 ◽  
Vol 74 (6) ◽  
pp. WCA83-WCA93 ◽  
Author(s):  
Naoshi Aoki ◽  
Gerard T. Schuster
Keyword(s):  
Least Squares ◽  
Model Test ◽  
Field Data ◽  
Waveform Inversion ◽  
Synthetic Data ◽  
Regularization Scheme ◽  
Practical Tool ◽  
Marine Data ◽  
The Cost ◽  
Least Squares Migration

Least-squares migration (LSM) is a linearized waveform inversion for estimating a subsurface reflectivity model that, relative to conventional migration, improves spatial resolution of migration images. The cost, however, is high because LSM typically requires 10 or more iterations, which is at least 20 times more than the CPU cost of conventional migration. To alleviate this expense, we offer a deblurring filter that can be used in a regularization scheme or a preconditioning scheme to give acceptable LSM images with less than one-third the cost of the standard LSM method. Our results in applying deblurred LSM to synthetic data and field data support this claim. In particular, a Marmousi2 model test shows that the data residual for preconditioned deblurred LSM decreases rapidly in the first iteration, which is equivalent to 10 or more iterations of LSM. Empirical results suggest that regularized deblurred LSM after three iterations is equivalent to about 10 iterations of LSM. Applying deblurred LSM to 2D marine data gives a higher-resolution image compared to those from migration or LSM with three iterations. These results suggest that LSM combined with a deblurring filter allows LSM to be a fast, practical tool for improved imaging of complicated structures.

Data-Driven Optimization of Intermittent Gas Production in Mature Fields Assisted by Deep Learning and a Population-Based Global Optimizer

2021 ◽  
Author(s):  
Javier Fatou Gómez ◽  
Pejman Shoeibi Omrani ◽  
Stefan Philip Christian Belfroid
Keyword(s):  
Deep Learning ◽  
Field Data ◽  
Synthetic Data ◽  
Population Based ◽  
Gas Production ◽  
Production Optimization ◽  
Global Optimizer ◽  
Single Well ◽  
Production Increase ◽  
Operational Criteria

Abstract In gas wells, decreased/unstable production can occur due to difficult-to-predict dynamic effects resulted from late-life phenomena, such as liquid loading and flooding. To minimize the negative impact of these effects, maximize production and extend the wells’ lifetime, wells are often operated in an intermittent production regime. The goal of this work is to find the optimum production and shut-in cycles to maximize intermittent gas production as a decision support to operators. A framework suitable for single and multiple wells was developed by coupling a Deep Learning forward model trained on historical data with a population-based global optimizer, Particle Swarm Optimization (PSO). The forward model predicts the production rates and wellhead pressure during production and shut-in conditions, respectively. The PSO algorithm optimizes the operational criteria given operational and environmental objectives, such as maximizing production, minimizing start-up/shut-in actions, penalizing emissions under several constraints such as planned maintenances and meeting a contract production value. The accuracy of the Deep Learning models was tested on synthetic and field data. On synthetic data, mature wells were tested under different reservoir conditions such as initial water saturation, permeability and flow regimes. The relative errors in the predicted total cumulative production ranged between 0.5 and 4.6% for synthetic data and 0.9% for field data. The mean errors for pressure prediction were of 2-3 bar. The optimization framework was benchmarked for production optimization and contract value matching for a single-well (on field data) and a cluster of wells (synthetic data). Single-well production optimization of a North Sea well achieved a 3% production increase, including planned maintenances. Production optimization for six wells resulted in a 21% production increase for a horizon of 30 days, while contract value matching yielded 29/30 values within 3% of the target. The most optimum, repeatable and computationally efficient results were obtained using critical pressure/gas flowrates as operational criteria. This could enable real-time gas production optimization and operational decision-making in a wide range of well conditions and operational requirements.

Sign in / Sign up

Export Citation Format

Share Document