Deep learning unflooding for robust subsalt waveform inversion

Alali, Abdullah A.; Kazei, Vladimir; Kalita, Mahesh; Alkhalifah, Tariq

doi:10.1111/1365-2478.13193

Cited by 11 publications

(7 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We adopt the 1D U-net used in Alali et al (2022) for all the three networks. It consists of four encoder/decoder blocks and a bottleneck that connects the two parts.…”

Section: The Neural Network Setupmentioning

confidence: 99%

“…Normally, implementing FWI is expensive, which makes it impractical for generating the necessary number of samples for the training dataset. However, applying FWI on 1D models is much cheaper as it can be implemented using only a single shot (Alali et al, 2022). The flooding and undlooding are naturally applied on the vertical dimension; thus, we implement a 1D FWI on randomly generated 1D models.…”

Section: Training Datasetmentioning

confidence: 99%

“…Then, we train U-net2 with similar hyper-parameters as U-net1. For the unflooding with U-net3, we use the same network as in Alali et al (2022) without further training as it is trained on the same setup.…”

Section: Training Datasetmentioning

confidence: 99%

“…However, a drawback of this approach is that it requires advanced high resolution imaging, which is often expensive. Alali et al (2022) used U-net to automatically unflood the salt and approximate the subsalt velocity in FWI framework only without applying any imaging by training U-net in a regressive regime. However, they assumed accurate pre-salt sediments and salt flooding.…”

Section: Introductionmentioning

confidence: 99%

“…Here, we extend Alali et al (2022) implementation by starting without any salt information and a poor initial model. Specifically, we apply FWI starting from a constant velocity model and build the salt body using three networks sequentially: two network for flooding the salt and one to unflood the salt to its base.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Integrating U-net with full-waveform inversion for an efficient salt body construction

Alaliand

Alkhalifah

2022

Second International Meeting for Applied Geoscience &Amp; Energy

Self Cite

View full text Add to dashboard Cite

Full waveform inversion (FWI) often faces a lot of difficulty inverting for the subsurface when starting with a poor initial model, especially in complex subsurface regions, such as those containing salt bodies. Thus, human intervention is often needed to pick the salt boundary and improve the starting model for the inversion, which can be erroneous and time consuming. Recently, machine learning has greatly assisted us in interpreting the salt body from seismic images, which helps automate the salt building process. However, such ML algorithms require seismic images, which in turn requires a reasonable good velocity model. Here, we build the salt body using low frequency FWI with the aid of neural networks, specifically U-net. The inversion is implemented in a multi-scale fashion and the networks for flooding and unflooding the salt are applied after each scale. We start the inversion from a constant velocity model using low frequencies of 3-7 Hz. Then, we apply a U-net to improve the inversion and flood the salt. We repeat this process of FWI and flooding in the next frequency bandwidth. In the last frequency bandwidth, we use a U-net for unflooding and impliment a final FWI. The networks were trained in a supervised manner using 1D inverted velocity models. We show the potential of the approach on the center part of BP 2004 salt model.

show abstract

“…We adopt the 1D U-net used in Alali et al (2022) for all the three networks. It consists of four encoder/decoder blocks and a bottleneck that connects the two parts.…”

Section: The Neural Network Setupmentioning

confidence: 99%

Section: Training Datasetmentioning

confidence: 99%

Section: Training Datasetmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Integrating U-net with full-waveform inversion for an efficient salt body construction

Alaliand

Alkhalifah

2022

Second International Meeting for Applied Geoscience &Amp; Energy

Self Cite

View full text Add to dashboard Cite

show abstract

A fine‐tuning workflow for automatic first‐break picking with deep learning

Mardan,

Blouin,

Fabien‐Ouellet

et al. 2024

Near Surface Geophysics

View full text Add to dashboard Cite

First‐break picking is an essential step in seismic data processing. For reliable results, first arrivals should be picked by an expert. This is a time‐consuming procedure and subjective to a certain degree, leading to different results for different operators. In this study, we have used a U‐Net architecture with residual blocks to perform automatic first‐break picking based on deep learning. Focusing on the effects of weight initialization on first‐break picking, we conduct this research by using the weights of a pre‐trained network that is used for object detection on the ImageNet dataset. The efficiency of the proposed method is tested on two real datasets. For both datasets, we pick manually the first breaks for less than 10 of the seismic shots. The pre‐trained network is fine‐tuned on the picked shots, and the rest of the shots are automatically picked by the neural network. It is shown that this strategy allows to reduce the size of the training set, requiring fine‐tuning with only a few picked shots per survey. Using random weights and more training epochs can lead to a lower training loss, but such a strategy leads to overfitting as the test error is higher than the one of the pre‐trained network. We also assess the possibility of using a general dataset by training a network with data from three different projects that are acquired with different equipment and at different locations. This study shows that if the general dataset is created carefully it can lead to more accurate first‐break picking; otherwise, the general dataset can decrease the accuracy. Focusing on near‐surface geophysics, we perform traveltime tomography and compare the inverted velocity models based on different first‐break picking methodologies. The results of the inversion show that the first breaks obtained by the pre‐trained network lead to a velocity model that is closer to the one obtained from the inversion of expert‐picked first breaks.

show abstract

Salt reconstruction in full-waveform inversion using topology optimization techniques

Gonçalves

Silva

2023

Geophysical Journal International

View full text Add to dashboard Cite

Summary With the increasing advances in the oil and gas industry, seismic imaging near or under salt structures has become an important point in deep-water exploration. Detailed velocity models of these areas are particularly interesting not only to characterize hydrocarbon reservoirs but also to identify potential sites for hydrogen and carbon dioxide storage in offshore salt caverns. Thus, we study the Full Waveform Inversion (FWI) for the salt reconstruction in acoustic media with constant density considering the time-harmonic wave propagation in a Finite Element formulation using the Topology Optimization (TO) method. This problem is challenging due to the strong velocity contrast between salt bodies and the sedimentary background, in addition to the lack of low-frequency data and the inherent ill-posedness of the inverse problem. In this context, we incorporate techniques from the TO field, usually used in design applications, to overcome or reduce these known problems. We initially defined the squared slowness as a combination of two fields, one related to the salt shape and the other to the background. An interpolation rule based on the Solid Isotropic Material with Penalization (SIMP) method, combined with filtering and projection schemes, is used to find the shape of the salt bodies with increased sharpness interfaces. A Helmholtz-type filter is applied to modify the gradient aimed to regularize the problem and provide a more stable way for the salt shape to evolve during the inversion process. In particular, we demonstrate that the proposed approach may be relevant for reconstructing media with salt bodies when a suitable starting model is unavailable, and sharp interfaces are required. In addition, we present inversion results from synthetic data generated by a variable density model to demonstrate the approach capability when subjected to a reconstruction application.

show abstract

Deep learning unflooding for robust subsalt waveform inversion

Abstract: This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as

Cited by 11 publications

References 46 publications

Integrating U-net with full-waveform inversion for an efficient salt body construction

Integrating U-net with full-waveform inversion for an efficient salt body construction

A fine‐tuning workflow for automatic first‐break picking with deep learning

Salt reconstruction in full-waveform inversion using topology optimization techniques

Contact Info

Product

Resources

About