A Convolutional Neural Network-Based Seismic Full Waveform Inversion for Coastal Subsurface Exploration

Moadeli, Amir Hossein; Danyali, Habibollah; Helfroush, Mohammad Sadegh; Rahnema, Hossein

doi:10.22034/ijcoe.2024.456865.1080

A Convolutional Neural Network-Based Seismic Full Waveform Inversion for Coastal Subsurface Exploration

Articles in Press

Document Type : Original Research Article

Authors

Amir Hossein Moadeli ¹

Habibollah Danyali ¹

Mohammad Sadegh Helfroush ¹

Hossein Rahnema ²

¹ Department of Electrical Engineering, Shiraz university of technology, Shiraz, Iran

² Department of Civil and Environmental Engineering, Shiraz University of Technology, Shiraz, Iran

10.22034/ijcoe.2024.456865.1080

Abstract

In this study, a convolutional neural network has been proposed to estimate the speed model of the subsurface structure based on multi-channel surface wave (MASW) analysis, addressing the Full-waveform Inversion (FWI) problem. A structure of deep supervision U-net architecture with two-stage encoder-decoder has been used in the proposed network. The proposed network structure is evaluated by 2D data models of an open source synthetic seismic Dataset, OPEN FWI, with different layering structures using Mean Squared Error (MSE), Structural Similarity (SSIM) and Mean Absolute Error (MAE) criteria. Also, in order to evaluate the robustness of the proposed method, the effect of filtering and removing parts of the input data have been investigated. A comparison results with several FWI methods in the current literature have been provided. The experimental results show that the proposed method is able to provide more accuracy in estimating and reconstructing the velocity model which results in better subsurface layering estimation.

Keywords

Seismic waves

Multi-channel surface wave analysis

Seismic inversion

deep learning

Full Waveform Inversion

Subjects