Survey of Deep Learning and Physics-Based Approaches in Computational Wave Imaging

Area of Science:

• Computational imaging
• Wave physics
• Data science

Background:

• Computational wave imaging (CWI) analyzes wave signals to reveal material structures and properties.
• Existing CWI methods are physics-based or deep learning-based.
• Physics-based methods offer accuracy but are computationally intensive and face ill-posedness challenges.

Purpose of the Study:

• To review the integration of scientific machine learning (ML), particularly deep neural networks, with traditional physics-based CWI methods.
• To provide a structured framework consolidating research across computational imaging, wave physics, and data science.
• To identify lessons learned, technical hurdles, and emerging trends in ML-enhanced CWI.

Main Methods:

• Systematic literature analysis of ML techniques applied to CWI problems.
• Consolidation of research spanning computational imaging, wave physics, and data science.
• Framework development for understanding ML integration in CWI.

Main Results:

• Deep learning techniques are being developed to enhance and integrate with physics-based CWI.
• A structured framework categorizes and consolidates diverse research efforts.
• Identified key lessons, challenges, and trends in ML-driven CWI.

Conclusions:

• The integration of ML with physics-based methods presents a promising direction for CWI.
• Addressing technical hurdles and leveraging emerging trends will further advance the field.
• This review offers insights for researchers in computational imaging, wave physics, and data science.