Generalization of learned Fourier-based phase-diversity wavefront sensing.

Area of Science:

• Optical Engineering
• Image Processing
• Computational Optics

Background:

• Accurate wavefront sensing is crucial for optical system performance.
• Traditional phase diversity wavefront sensing (PDWS) methods can be sensitive to initialization, risking local minima.
• Existing neural network approaches for PDWS often lack generalization due to reliance on specific training parameters.

Purpose of the Study:

• To develop a generalized Fourier-based PDWS method that overcomes the limitations of training-dependent networks.
• To improve the robustness and applicability of neural network-based PDWS across diverse imaging conditions.
• To achieve accurate aberration estimation independent of specific object and optical system parameters.

Main Methods:

• Proposed a generalized Fourier-based PDWS method combining an object-independent neural network with system-independent image processing.
• Utilized low-frequency Fourier domain coefficients for aberration estimation.
• Trained a single network and tested its performance across multiple, varied optical settings.

Main Results:

• Demonstrated that a network trained with one setting can be successfully applied to images from four other distinct settings.
• Achieved high accuracy in estimating aberrations, with mean RMS residual errors as low as 0.032 λ.
• Reported that 98.9% of RMS residual errors were below 0.05 λ for aberrations within [0.2 λ, 0.4 λ].

Conclusions:

• The proposed generalized Fourier-based PDWS method offers robust and accurate wavefront aberration estimation.
• The object-independent network significantly enhances the generalization ability of PDWS systems.
• This approach provides a more versatile and reliable solution for wavefront sensing in various optical applications.