Direct phase sensing via an end-to-end model under high-order aberrations

Area of Science:

• Optical engineering
• Image processing
• Deep learning

Background:

• Wavefront estimation is crucial for optical system performance.
• Traditional methods like modal reconstruction struggle with high-order aberrations.
• Phase diversity imaging provides data for wavefront reconstruction.

Purpose of the Study:

• To develop an end-to-end deep learning model for direct exit pupil wavefront estimation.
• To improve wavefront reconstruction accuracy, particularly for high-order aberrations.
• To enhance the quality of reconstructed images from degraded optical systems.

Main Methods:

• An end-to-end deep learning model was proposed to estimate the exit pupil wavefront directly from phase diversity images.
• Zonal reconstruction was employed, avoiding the mode restrictions inherent in modal reconstruction (e.g., ResNet50 with Zernike modes [3,28]).
• Simulated experimental data were used to evaluate the performance against modal reconstruction.

Main Results:

• Zonal reconstruction significantly outperformed modal reconstruction in restoring high-order aberrations.
• The deep learning model achieved higher fitting accuracy across all aberrations without mode restrictions.
• Reconstructed images showed richer high spatial frequency details and improved accuracy.

Conclusions:

• Zonal reconstruction, implemented via deep learning, is a superior method for exit pupil wavefront estimation compared to modal reconstruction.
• The proposed method effectively restores high-order aberrations, leading to better image reconstruction quality.
• This approach offers a promising solution for improving optical system performance and image fidelity.