Few-shot real-time quantitative phase imaging based on lightweight networks.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Image Processing

Background:

• Quantitative phase imaging (QPI) is vital for environmental monitoring and atmospheric turbulence analysis.
• Current deep learning QPI methods struggle with high computational costs and large datasets, hindering real-time applications.
• Millisecond-scale processing is essential for demanding QPI applications.

Purpose of the Study:

• To develop a few-shot, real-time quantitative phase imaging algorithm.
• To overcome the limitations of computational complexity and data requirements in existing deep learning QPI methods.
• To enable efficient and reliable phase reconstruction in data-scarce scenarios.

Main Methods:

• Proposed a few-shot real-time quantitative phase imaging algorithm.
• Introduced a novel, lightweight StarNet architecture for efficient feature extraction and dimensional expansion.
• Focused on reducing computational overhead for millisecond-scale processing.

Main Results:

• Achieved millisecond-scale phase reconstruction per frame.
• Demonstrated effective performance even with low signal-to-noise ratios.
• Showcased robustness against diverse random phase disturbances.
• Validated the algorithm's efficiency in data-scarce environments.

Conclusions:

• The developed algorithm significantly reduces computational overhead for real-time QPI.
• It enables practical deployment of real-time QPI technology.
• Offers an efficient and reliable solution for multi-frame inverse imaging problems with limited data.