Experimental wavefront sensing techniques based on deep learning models using a Hartmann-Shack sensor for visual...

Area of Science:

• Visual Optics
• Optical Engineering
• Artificial Intelligence in Optics

Background:

• Wavefront sensing is critical for assessing optical quality in systems like the human eye.
• Traditional Hartmann-Shack wavefront sensors (HSS) have limitations in precision, dynamic range, and speed.
• Deep learning offers potential solutions to overcome existing wavefront sensing challenges.

Purpose of the Study:

• To develop a novel approach to enhance Hartmann-Shack wavefront sensor (HSS) performance using deep learning.
• To investigate the efficacy of a modified ResNet convolutional neural network (CNN) for wavefront aberration reconstruction.
• To evaluate the impact of the proposed CNN model on accuracy, speed, and dynamic range compared to traditional methods.

Main Methods:

• A modified ResNet convolutional neural network (CNN) architecture was developed for wavefront sensing.
• Experimental datasets were generated using a custom monocular visual simulator, including noise-free and speckle noise conditions.
• The CNN model was trained and tested on images from the Hartmann-Shack wavefront sensor (HSS).

Main Results:

• The proposed CNN model demonstrated superior accuracy in processing HSS images.
• Wavefront aberration reconstruction time was reduced by 300% to 400% compared to traditional methods.
• The dynamic range of wavefront sensing was increased by 315.6% using the CNN approach.

Conclusions:

• The modified ResNet CNN model significantly enhances wavefront sensing capabilities.
• This deep learning approach offers a practical and efficient solution for improving optical quality evaluation.
• The findings have implications for advancing applications in visual optics and related fields.