Related Experiment Video
Updated: Jul 28, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
588
Sensing prior constraints in deep neural networks for solving exploration geophysical problems
Xinming Wu1,2, Jianwei Ma3, Xu Si1,2
1School of Earth and Space Sciences, University of Science and Technology of China, Hefei, 230026 China.
Summary
This study introduces three methods to improve deep neural networks (DNNs) for geophysics by incorporating domain knowledge. These strategies enhance the accuracy and reliability of subsurface characterization using geophysical data.
Area of Science:
- Geophysics
- Data Science
- Machine Learning
Background:
- Geophysical data analysis is crucial for subsurface characterization.
- Deep learning methods offer potential but face challenges like poor generalizability and interpretability.
Purpose of the Study:
- To address challenges in data-driven deep learning for geophysics.
- To present strategies for imposing domain knowledge constraints on deep neural networks (DNNs).
Main Methods:
- Integrating constraints into data via synthetic datasets from forward modeling.
- Designing custom DNN layers with physical operators for knowledge consistency.
- Implementing prior information and laws as regularization terms in loss functions.
Main Results:
- Demonstrated effectiveness of the three strategies in geophysical applications.
- Improved generalizability, interpretability, and physical consistency of DNNs.
Conclusions:
- Domain knowledge integration significantly enhances DNN performance in geophysics.
- These strategies are applicable to various geophysical tasks including processing, imaging, and model building.
