Deep learning seismic substructure detection using the frozen Gaussian approximation

DOI10.1016/J.JCP.2020.109313zbMATH Open1435.86002arXiv1810.06610OpenAlexW3006486097MaRDI QIDQ778276FDOQ778276

Authors: James C. Hateley, Jay Roberts, Kyle Mylonakis, Xu Yang 

Publication date: 2 July 2020

Published in: Journal of Computational Physics (Search for Journal in Brave)

Abstract: We propose a deep learning algorithm for seismic interface and pocket detection with neural networks trained by synthetic high-frequency displacement data efficiently generated by the frozen Gaussian approximation (FGA). In seismic imaging high-frequency data is advantageous since it can provide high resolution of substructures. However, generation of sufficient synthetic high-frequency data sets for training neural networks is computationally challenging. This bottleneck is overcome by a highly scalable computational platform built upon the FGA, which comes from the semiclassical theory and approximates the wavefields by a sum of fixed-width (frozen) Gaussian wave packets. Data is generated from a forward simulation of the elastic wave equation using the FGA. This data contains accurate traveltime information (from the ray path) but not exact amplitude information (with asymptotic errors not shrinking to zero even at extremely fine numerical resolution). Using this data we build convolutional neural network models using an open source API, GeoSeg, developed using Keras and Tensorflow. On a simple model, networks, despite only being trained on FGA data, can detect an interface with a high success rate from displacement data generated by the spectral element method. Benchmark tests are done for P-waves (acoustic) and P- and S-waves (elastic) generated using the FGA and a spectral element method. Further, results with a high accuracy are shown for more complicated geometries including a three layered model, and a 2D-pocket model where the neural networks trained by both clean and noisy data.

Full work available at URL: https://arxiv.org/abs/1810.06610

Recommendations

• Seismic stratum segmentation using an encoder-decoder convolutional neural network
• Seismic inversion with deep learning. A proposal for litho-type classification
• Fast geostatistical seismic inversion coupling machine learning and Fourier decomposition
• Frozen Gaussian approximation for 3D seismic tomography
• Stochastic seismic waveform inversion using generative adversarial networks as a geological prior
• Modeling conditional distributions of facies from seismic using neural nets
• Seismic wave propagation modeling on CNN-UM architecture
• An encoder-decoder deep surrogate for reverse time migration in seismic imaging under uncertainty
• The use of convolutional neural networks to search for fractures in the geological media

zbMATH Keywords

image segmentationfrozen Gaussian approximationelastic wave equationseismic tomographyconvolutional neural networkhigh-frequency wavefield

Mathematics Subject Classification ID

Random fields; image analysis (62M40) Neural nets and related approaches to inference from stochastic processes (62M45) Seismology (including tsunami modeling), earthquakes (86A15)

Cites Work

• Eulerian Gaussian beam method for high frequency wave propagation in heterogeneous media with discontinuities in one direction
• Frozen Gaussian approximation for general linear strictly hyperbolic systems: formulation and Eulerian methods

Cited In (5)

• Deep learning for ranking response surfaces with applications to optimal stopping problems
• Towards blending physics-based numerical simulations and seismic databases using generative adversarial network
• Frozen Gaussian approximation for 3D seismic tomography
• Explainable time-frequency convolutional neural network for microseismic waveform classification
• Seismic inversion with deep learning. A proposal for litho-type classification

Uses Software

• U-Net

This page was built for publication: Deep learning seismic substructure detection using the frozen Gaussian approximation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q778276)