Abstract:This paper discusses the joint inversion of seismic and electromagnetic data in geophysical exploration. It recognizes seismic exploration as a primary method, especially through full waveform inversion (FWI) that enables high-resolution subsurface velocity estimation. However, it acknowledges the severe non-uniqueness problem associated with inverting subsurface velocities using solely seismic data. The abstract introduces electromagnetic exploration as an effective supplement due to its expansive detection range, cost-effectiveness, sensitivity to subsurface anomalies, and provision of deep resistivity information. By jointly inverting electromagnetic and seismic data, complementary information from both datasets is integrated, ameliorating the non-uniqueness issue and enhancing the accuracy and reliability of the inversion results. However, existing joint inversion techniques encounter hurdles such as complex objective function design, difficulty in achieving convergence, and inadequate coupling between seismic and electromagnetic data. To surmount these challenges, we propose a Multi-scale Feature Fusion Network (MFF Net) grounded in joint learning. This approach harnesses the robust nonlinear fitting capacities of neural networks to optimize the objective function efficiently. Furthermore, coupling between seismic and electromagnetic data is established across multiple sampling scales. By leveraging the frequency band complementarity of pre-stack seismic and electromagnetic data and considering the low-frequency constraints from electromagnetic data alongside mid-to-high frequency constraints from seismic data, high-resolution resistivity and velocity models are inverted. This ultimately boosts the precision and reliability of the inversion results. We validate the efficacy of this approach using synthetic data, demonstrating the effective reconstruction of the underground salt dome structure, even in the absence of low-frequency data.

Global-Feature-Fusion and Multiscale Network for Low-Frequency Extrapolation

Transformer For Low-frequency Extrapolating of Seismic Data

MFF Net: A Multiscale Feature Fusion Network for Electromagnetic and Seismic Joint Inversion

Low-Frequency Extrapolation of Prestack Viscoacoustic Seismic Data Based on Dense Convolutional Network

Extrapolated full-waveform inversion with deep learning

A self-supervised learning framework for seismic low-frequency extrapolation

Layered low-frequency extrapolation with deep learning in full-waveform inversion

HGM: A General High-Order Spatial and Spectral Global Feature Fusion Module for Visual Multitasking

A Novel AMSS-FFN for Underwater Multisource Localization Using Artificial Lateral Line.

Multiscale Data-driven Seismic Full-waveform Inversion with Field Data Study

Efficient progressive transfer learning for full waveform inversion with extrapolated low-frequency reflection seismic data

Low-frequency constrained seismic impedance inversion combining large kernel attention and long short-term memory

DD-Net: Dual Decoder Network With Curriculum Learning for Full Waveform Inversion

MFFnet: A Seismic Phase Picking Network Based on Multiple Feature Fusion

Multitask Seismic Inversion Based on Deformable Convolution and Generative Adversarial Network

Full-Waveform Inversion of Multifrequency GPR Data Using a Multiscale Approach Based on Deep Learning

Multiscale Fusion Network Based on Global Weighting for Hyperspectral Feature Selection

Deep Learning-Based Low-Frequency Extrapolation and Impedance Inversion of Seismic Data

Inversion-DeepONet: A Novel DeepONet-Based Network with Encoder-Decoder for Full Waveform Inversion

An Augmented Lagrangian Method-Based Deep Iterative Unrolling Network for Seismic Full-Waveform Inversion

Multichannel deconvolution with a high-frequency structural regularization