Deep learning electromagnetic inversion with convolutional neural networks

Abstract

Geophysical inversion attempts to estimate the distribution of physical properties in the Earth's interior from observations collected at or above the surface. Inverse problems are commonly posed as least-squares optimization problems in highdimensional parameter spaces. Existing approaches are largely based on deterministic gradient-based methods, which are limited by nonlinearity and nonuniqueness of the inverse problem. Probabilistic inversion methods, despite their great potential in uncertainty quantification, still remain a formidable computational task. In this paper, I explore the potential of deep learning methods for electromagnetic inversion. This approach does not require calculation of the gradient and provides results instantaneously. Deep neural networks based on fully convolutional architecture are trained on large synthetic datasets obtained by full 3-D simulations. The performance of the method is demonstrated on models of strong practical relevance representing an onshore controlled source electromagnetic CO 2 monitoring scenario. The pre-trained networks can reliably estimate the position and lateral dimensions of the anomalies, as well as their resistivity properties. Several fully convolutional network architectures are compared in terms of their accuracy, generalization, and cost of training. Examples with different survey geometry and noise levels confirm the feasibility of the deep learning inversion, opening the possibility to estimate the subsurface resistivity distribution in real time.

Deep neural networks for 1D impedance inversion

“…Deep CNNs with many stacked layers are highly efficient in processing images and data with a grid-like topology, which made these neural networks widely used in computer vision applications (e.g., Krizhevsky et al 2012;Szegedy et al 2015). The fully convolutional architecture can be efficiently used in multi-dimensional inversion (Puzyrev 2018), where the decoder is used to output an estimation of the spatial distribution of the subsurface properties.…”

Error control and loss functions for the deep learning inversion of borehole resistivity measurements

“…This work does not discuss optimal data sampling techniques nor the decision-making for the optimal selection of DNN architectures. 34,35 Those subjects are possible future work. However, for this article to be self-contained, we briefly describe in Appendix the architecture of the DNN we use.…”

“…Then, we apply the resulting DNN approximations to synthetic examples, which help us elucidate their main advantages and limitations. This work does not discuss optimal data sampling techniques nor the decision‐making for the optimal selection of DNN architectures 34,35 . Those subjects are possible future work.…”

Combining discrete cosine transform and convolutional neural networks to speed up the Hamiltonian Monte Carlo inversion of pre‐stack seismic data

“…In geophysics, CNNs have initially been applied to aid structural interpretation of geophysical data such as seismic horizon and fault interpretation, and seismic texture identification (Xiong et al ., 2018; Waldeland et al ., 2018). Then, they have been extended to quantitatively solve geophysical problems such as velocity estimation (Araya‐Polo et al ., 2018), full‐waveform inversion (Lewis and Vigh, 2017; Richardson, 2018; Wu and McMechan, 2019; Yang and Ma, 2019), impedance inversion (Das et al ., 2019), electromagnetic inversion (Puzyrev, 2019), lithology prediction (Raeesi et al ., 2012; Hall, 2016), seismic deblending (Sun et al ., 2020), missing trace interpolation and noise attenuation (Liu et al ., 2018; Wang et al ., 2019; Mandelli et al ., 2019), multiple attenuation (Ma, 2018), pre‐stack seismic waveform classification and first‐break picking (Yuan et al ., 2018), automatic velocity analysis (Park and Sacchi, 2020) and reservoir characterization studies (Zhong et al ., 2019). In particular, Zhang et al .…”