Optimizing delamination imaging via full wavefield segmentation using augmented simulated wavefield data

Yitian Yan

doi:10.1016/j.ultras.2025.107789

Optimizing delamination imaging via full wavefield segmentation using augmented simulated wavefield data

2025-08-13

0

PRE

AI

DOI：10.1016/j.ultras.2025.107789

Original

Original request for help

Save

Abstract

En 中文

• A data preprocessing strategy is proposed to improve the performance of DL models in full wavefield segmentation for delamination imaging. By applying this strategy, models trained exclusively on simulated data can be effectively generalized to experimental scenarios, enabling accurate delamination imaging. • Derived GW modes caused by delamination are extracted from the full wavefields for model training. This approach compresses 3D raw full wavefield data into 2D frequency domain feature spectrums while highlighting the energy response induced by delamination. • A data augmentation strategy that combines symmetric flipping and the addition of white Gaussian noise is employed to improve the generalization of the delamination imaging models, enabling them to better handle real-world measurement conditions. • Seven distinct DL models are employed and evaluated, showcasing the effectiveness and benefits of the data preprocessing strategy.

Keywords:

delamination imaging

deep learning

data preprocessing

full wavefield segmentation

data augmentation

Optimizing delamination imaging via full wavefield segmentation using augmented simulated wavefield data

Abstract

Journal

Organization

Cited Papers