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 high-dimensional 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 CO2 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.

Microwave Data Inversion With a Model Compression Scheme Based on Deep Learning

A Nonlinear Model Compression Scheme Based on Variational Autoencoder for Microwave Data Inversion

Enhanced Deep Learning Approach Based on the Deep Convolutional Encoder–Decoder Architecture for Electromagnetic Inverse Scattering Problems

Microwave Inversion For Sparse Data Using Descent Learning Technique

Microwave Imaging by Deep Learning Network: Feasibility and Training Method

Microwave Imaging under Oblique Illumination

Application of the model-based inversion algorithm to microwave data

Controlled-source electromagnetic data inversion using the model compression scheme

Three Dimensional Microwave Data Inversion in Feature Space for Stroke Imaging.

A deep learning enhanced inverse scattering framework for microwave imaging of piece-wise homogeneous targets

Pixel- and Model-Based Microwave Inversion with Supervised Descent Method for Dielectric Targets

Reconstruction of 3D targets from microwave measurements using a model-based inversion scheme

A Three-Dimensional Model-Based Inversion Algorithm Using Radial Basis Functions for Microwave Data

Numerical Analysis of Inverse Scattering in Microwave Imaging

Semantic-Electromagnetic Inversion With Pretrained Multimodal Generative Model

Contrast Source Inversion Algorithm Applied in 3-D Microwave Imaging

MDLI-Net: Model-Driven Learning Imaging Network for High-Resolution Microwave Imaging with Large Rotating Angle and Sparse Sampling

A sparse coding approach to inverse problems with application to microwave tomography

Learning-Based Quantitative Microwave Imaging With a Hybrid Input Scheme

Fast Full-wave Microwave Imaging with Physics Embedded Deep Neural Network

Deep learning electromagnetic inversion with convolutional neural networks