Large-scale piston error detection technology for segmented optical mirrors via convolutional neural networks.

Area of Science:

• Optical engineering
• Machine learning applications
• Wavefront sensing

Background:

• Segmented optical mirrors require precise alignment (cophasing) for optimal performance.
• Traditional wavefront sensors like Shack-Hartmann struggle with submirror piston errors and large piston ranges.
• Existing solutions to improve piston error detection often increase system complexity and cost.

Purpose of the Study:

• To develop a novel method for detecting submirror piston errors in segmented optical systems.
• To overcome the limitations of existing wavefront sensing techniques for piston error detection.
• To reduce the complexity and manufacturing cost associated with cophasing segmented mirrors.

Main Methods:

• Utilized a convolutional neural network (CNN) to classify piston error ranges.
• Constructed a target-independent feature vector using in-focus and defocused images.
• Employed multi-wavelength imaging to extend the detection range beyond traditional limits.

Main Results:

• The CNN effectively distinguished piston error ranges for individual submirrors.
• The feature vector approach eliminated dependence on specific imaging targets.
• The method demonstrated successful piston error detection across a wider range, surpassing fundamental limits.

Conclusions:

• The proposed CNN-based method offers an effective solution for piston error detection in segmented optical mirrors.
• This approach enhances cophasing accuracy while avoiding the need for additional complex optical components.
• The technique presents a cost-effective and robust alternative for improving optical system alignment.