Automated interpretation of top and base salt using deep-convolutional networks

Gramstad, Oddgeir; Nickel, Michael

doi:10.1190/segam2018-2996306.1

Cited by 25 publications

(6 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Considering the physical property differences among the salt and the surrounding sediment layers, methods to classify salt structure boundaries have been adopted e.g., seismic attribution extraction (Di et al, 2019a;2019b). Salt-structure detection has been aided by machinelearning methods development, including normalized full gradient machines (Soleimani et al, 2018), the oriented gradients histogram combined with support vector machines (Hosseini-Fard et al, 2022) and mostly, convolution neural networks (CNNs) (Di et al, 2018;Gramstad and Nickel, 2018;Karchevskiy et al, 2018). However, these methods require a considerable amount of seismic data to calculate the seismic attributes.…”

Section: Introductionmentioning

confidence: 99%

Salt structure identification based on U-net model with target flip, multiple distillation and self-distillation methods

Jin²,

Xia³

et al. 2023

Front. Earth Sci.

View full text Add to dashboard Cite

Salt structures are crucial targets in oil and gas seismic exploitation so that one fast, automatic and accurate method is necessary for accelerating salt structure identification in the exploitation process. With the development of machine-learning algorithms, geophysical scientists adopt machine-learning models to solve problems. Most machine-learning models in geophysics require mass data in the model training. However, the number of seismic images is limited and the class-imbalance is often existed in actuality, causing the machine-learning algorithms to be difficult to apply in exploitation projects. To overcome the challenge of the seismic images’ volume, this work collects a two-dimensional (2D) seismic images dataset and trains several U-net models with the methods of inversion and multiple distillation. Moreover, self-distillation is introduced to boost the model’s performance. A test using a public seismic dataset and the case of salt detection in the Hith evaporite in southern United Arab Emirates and western Oman shows the distillation method is able to identify salt structures automatically and accurately, which has great potential for application in actual exploitation.

show abstract

Section: Introductionmentioning

confidence: 99%

Salt structure identification based on U-net model with target flip, multiple distillation and self-distillation methods

Jin²,

Xia³

et al. 2023

Front. Earth Sci.

View full text Add to dashboard Cite

show abstract

“…A specific CNN architecture, known as U-net (Ronneberger et al, 2015), has achieved high accuracy compared to other architectures in salt detection (Shi et al, 2019;Sen et al, 2020;Naeini et al, 2020). Gramstad and Nickel (2018) proposed to train two CNNs: one to detect the top of the salt for flooding and the other to detect the base for unflooding. Even with these advancements, salt detection during the velocity model building stage is seldom easy due to noise and inaccurate positioning of the reflectors, thus, Zhao et al (2021) used U-net multiple times in an iterative manner between FWI and reverse time migration (RTM) images.…”

Section: Introductionmentioning

confidence: 99%

“…The first approach automates the interpretation in the top-down workflow. Gramstad and Nickel (2018) suggested training two individual convolutional networks (CNNs) to pick the top and the base of the salt. Sen et al (2020) shows that CNNs can detect a good initial salt model from noisy low-resolution seismic images.…”

Section: Introductionmentioning

confidence: 99%

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

Alaliand

Alkhalifah

2022

Second International Meeting for Applied Geoscience &Amp; Energy

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

“…Currently, there is a rapidly expanding number of scientific works developing and or applying CNNs and other machine learning techniques to assist researchers in detecting, modeling, or predicting specific features in a broad range of domains, including medical sciences (e.g., Ronneberger et al., 2015; Shen et al., 2017), remote sensing (e.g., Lary et al., 2016; Maggiori et al., 2017; Tasar, Tarabalka, & Alliez, 2019), chemistry (e.g., Schütt et al., 2017), climate sciences (e.g., Rolnick et al., 2019), engineering (e.g., Drouyer, 2020), and geosciences (e.g., Bergen et al., 2019). In the domain of geosciences, machine learning methods have especially flourished in earthquake seismology (e.g., Hulbert, Rouet‐Leduc, Jolivet, & Johnson, 2020; Kong et al., 2019; Mousavi & Beroza, 2020; Perol et al., 2018; Ross, Meier, & Hauksson, 2018; Ross, Meier, Hauksson, & Heaton, 2018; Rouet‐Leduc, Hulbert, & Johnson, 2019; Rouet‐Leduc, Hulbert, McBrearty, & Johnson, 2020; Zhu & Beroza, 2019), and seismic and geophysical imaging of the Earth's interior with a major focus on sub‐surface image interpretation for resource exploration (e.g., Babakhin et al., 2019; Dramsch & Lüthje, 2018; Gramstad & Nickel, 2018; Haber et al., 2019; Meier et al., 2007; Waldeland et al., 2018; Wu & Zhang, 2018). A few other works have been conducted in oceanography (Bézenac et al., 2019), geodesy (Rouet‐Leduc, Hulbert, McBrearty, & Johnson, 2020), volcanology (Ren, Peltier, et al., 2020), and rock physics (Hulbert, Rouet‐Leduc, Johnson, et al., 2019; Ren, Dorostkar, et al., 2019; Rouet‐Leduc, Hulbert, Lubbers, et al., 2017; Srinivasan et al., 2018; You et al., 2020).…”

Section: Introductionmentioning

confidence: 99%

Automatic Fault Mapping in Remote Optical Images and Topographic Data With Deep Learning

Mattéo

Manighetti

Tarabalka³

et al. 2021

JGR Solid Earth

View full text Add to dashboard Cite

Faults form dense, complex multi‐scale networks generally featuring a master fault and myriads of smaller‐scale faults and fractures off its trace, often referred to as damage. Quantification of the architecture of these complex networks is critical to understanding fault and earthquake mechanics. Commonly, faults are mapped manually in the field or from optical images and topographic data through the recognition of the specific curvilinear traces they form at the ground surface. However, manual mapping is time‐consuming, which limits our capacity to produce complete representations and measurements of the fault networks. To overcome this problem, we have adopted a machine learning approach, namely a U‐Net Convolutional Neural Network (CNN), to automate the identification and mapping of fractures and faults in optical images and topographic data. Intentionally, we trained the CNN with a moderate amount of manually created fracture and fault maps of low resolution and basic quality, extracted from one type of optical images (standard camera photographs of the ground surface). Based on a number of performance tests, we select the best performing model, MRef, and demonstrate its capacity to predict fractures and faults accurately in image data of various types and resolutions (ground photographs, drone and satellite images and topographic data). MRef exhibits good generalization capacities, making it a viable tool for fast and accurate mapping of fracture and fault networks in image and topographic data. The MRef model can thus be used to analyze fault organization, geometry, and statistics at various scales, key information to understand fault and earthquake mechanics.

show abstract

Automated interpretation of top and base salt using deep-convolutional networks

Cited by 25 publications

References 4 publications

Salt structure identification based on U-net model with target flip, multiple distillation and self-distillation methods

Salt structure identification based on U-net model with target flip, multiple distillation and self-distillation methods

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

Automatic Fault Mapping in Remote Optical Images and Topographic Data With Deep Learning

Contact Info

Product

Resources

About