Object-independent piston diagnosing approach for segmented optical mirrors via deep convolutional neural network

Area of Science:

• Optical Engineering
• Computational Optics
• Machine Learning Applications

Background:

• Piston diagnosing in optical systems often relies on target-dependent imaging, which is susceptible to submirror interference.
• Existing neural network methods face limitations due to their dependence on specific imaging targets and interference from optical components.

Purpose of the Study:

• To develop an object-independent piston diagnosing method for segmented mirrors.
• To overcome the challenges posed by submirror interference in piston error detection.
• To enhance the accuracy and generalizability of piston diagnosing techniques.

Main Methods:

• Generation of an object-independent feature image dataset for training.
• Construction of an 18-layer deep convolutional neural network (CNN).
• Creation of a dataset with 9600 images per submirror, incorporating sensitive area extraction.

Main Results:

• Achieved an average root mean square error of approximately 0.0622λ across six submirrors.
• Demonstrated superior CNN training performance with object-independent datasets containing effective details.
• Validated generalization ability through analysis of results diversity among submirrors.

Conclusions:

• The proposed method effectively diagnoses piston errors in segmented mirrors, mitigating submirror interference.
• The object-independent dataset and deep CNN approach require fewer images and offer improved training.
• This hardware-independent, fast, and widely applicable method can significantly advance segmented mirror piston diagnosing.