Learning velocity model for complex media with deep convolutional neural networks

TL;DR

Problem: recovering velocity distributions in complex media from boundary measurements. Approach: grid-characteristic forward modeling paired with a CNN-based inverse mapper (UNet), augmented by Fourier-domain input channels, Sobel-filter regularization, and ensemble averaging. Key findings: achieves a structure similarity index $SSIM$ of $0.93\\pm0.01$ on an open dataset of 1600 velocity models, with Fourier inputs providing significant gains and ensembling smoothing predictions; Sobel-based regularization is not essential. Significance: real-time capable, generalizable to non-destructive testing and biomedical imaging, and accompanied by an open 1600-model dataset for benchmarking.

Abstract

The paper considers the problem of velocity model acquisition for a complex media based on boundary measurements. The acoustic model is used to describe the media. We used an open-source dataset of velocity distributions to compare the presented results with the previous works directly. Forward modeling is performed using the grid-characteristic numerical method. The inverse problem is solved using deep convolutional neural networks. Modifications for a baseline UNet architecture are proposed to improve both structural similarity index measure quantitative correspondence of the velocity profiles with the ground truth. We evaluate our enhancements and demonstrate the statistical significance of the results.

Figures (6)

• Figure 1: Sample velocity models
• Figure 2: Sample wave propagation, emitter located in the center of the domain
• Figure 3: Sample wave propagation, emitter located in the left part of the domain
• Figure 4: Distributions of SSIM values for bootstrapped model training
• Figure 5: Prediction results for different network architectures