Minimum entropy-neural network approach to turbulent-image reconstruction

Area of Science:

• Optics and Photonics
• Computational Imaging
• Astrophysics

Background:

• Atmospheric turbulence distorts optical images, limiting clarity.
• Traditional methods struggle with rapid and complete turbulence correction.
• Understanding turbulence requires analyzing point-spread functions.

Purpose of the Study:

• To develop a neural network-based algorithm for enhanced imaging through atmospheric turbulence.
• To introduce and validate the Fourier division approach for image restoration.
• To assess the potential for turbulence characterization using the developed method.

Main Methods:

• Utilizing a standard model of optical turbulence where point-spread functions are random speckle superpositions.
• Implementing the Fourier division approach, requiring two rapid short-exposure images.
• Processing images via Fourier space division and a minimum entropy-neural net for point-spread function estimation.
• Applying inverse filtering to corrected images and averaging for a final sharp output.

Main Results:

• Computer simulations demonstrate significant removal of turbulence degradation.
• The method achieves high-quality image restoration in tens of seconds.
• Successful estimation of short-exposure point-spread functions.

Conclusions:

• The neural net-based Fourier division approach shows significant promise for real-time atmospheric turbulence mitigation.
• The method can produce sharp, turbulence-free images.
• Estimated point-spread functions offer insights into turbulence state and wind shear detection.