A theory-guided deep learning formulation of seismic waveform inversion

Sun, Jian; Niu, Zhan; Innanen, Kristopher A.; Li, Junxiao; Trad, Daniel

doi:10.1190/segam2019-3215762.1

Cited by 9 publications

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Sun et al. (2019); Sun, Niu, et al. (2020) proposed a theory‐guided seismic inversion framework in which the forward modeling is devised in a framework of recurrent neural network (RNN) and the inversion process is described as the RNN training using automatic differentiation that can be easily accelerated by deep learning platforms.…”

Section: Introductionmentioning

confidence: 99%

“…With this in mind, several ways of incorporating FWI with DNNs have been explored. Sun et al (2019); Sun, Niu, et al (2020) proposed a theory-guided seismic inversion framework in which the forward modeling is devised in a framework of recurrent neural network (RNN) and the inversion process is described as the RNN training using automatic differentiation that can be easily accelerated by deep learning platforms. The theory-guided RNN framework can also be extended to more complete multidimensional parameter estimations (Zhang et al, 2020) or electromagnetic inversion (Hu et al, 2021).…”

mentioning

confidence: 99%

See 1 more Smart Citation

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

et al. 2023

Self Cite

View full text Add to dashboard Cite

Full waveform inversion (FWI) is arguably the current state‐of‐the‐art amongst methodologies for imaging subsurface structures and physical parameters with seismic data; however, important challenges are faced in its implementation and use. Keys amongst these are (a) building a suitable initial model, from which a local minimum is unlikely to be reached, and (b) availability of tools for evaluation of uncertainty. An algorithm we refer to as implicit full waveform inversion (IFWI), designed using continuously and implicitly defined deep neural representations, appears in principle to address both of these issues. We observe in IFWI, with its random initialization and deep learning optimization, improved convergence relative to standard FWI model initialization and optimization. Models close to the global minimum, capturing relatively high‐resolution subsurface structures, are obtained. In addition, uncertainty analysis, though not solved in IFWI, is meaningfully addressed by approximating Bayesian inference with the addition of dropout neurons. Numerical experimentation with a range of 2D geological models is suggestive that IFWI exhibits a strong capacity for generalization, and is likely well‐suited for multi‐scale joint geophysical inversion.

show abstract

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

et al. 2023

Self Cite

View full text Add to dashboard Cite

show abstract

Full waveform inversion guided wave tomography with a recurrent neural network

Wang

Xiao

et al. 2023

Ultrasonics

View full text Add to dashboard Cite

Quality classification and inversion of receiver functions using convolutional neural network

Gan

Huang

et al. 2022

Geophysical Journal International

View full text Add to dashboard Cite

SUMMARY Convolutional neural network (CNN) is presented to implement quick quality classification and inversion for teleseismic P-wave receiver functions (RF). For the first case, a CNN is trained using field measured RFs from NE margin of the Tibetan Plateau to efficiently predict the quality of each input waveform. Signal-to-noise ratio and correlation are introduced to quantitatively determine the quality label of RF, avoiding the subjectivity of manual labelling. The trained network reduces the time needed for data processing and has higher accuracy and efficiency than conventional methods. Its good performance is confirmed by comparing it with manually selected data from NE of the Tibetan Plateau. The second case is an example of joint inverting teleseismic P-wave RF and surface wave dispersions for the estimation of earth S-wave structure and associated uncertainties. We train a UNet based on synthetic global Crust 5.1 models and standard earth models, as well as associated perturbed models to ensure enough generalization capacity. We find that the UNet inversion is robust and has a better performance to reconstruct subsurface ${V}_s$ distributions than the damping least-squares method, but at the expense of slightly higher data misfits. The pre-trained network can predict subsurface ${V}_s$ models and associated uncertainties beneath NE of the Tibetan Plateau, which is consistent with the published models.

show abstract

A theory-guided deep learning formulation of seismic waveform inversion

Cited by 9 publications

References 17 publications

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

Full waveform inversion guided wave tomography with a recurrent neural network

Quality classification and inversion of receiver functions using convolutional neural network

Contact Info

Product

Resources

About