Automated generation of easy-assembly off-axis three-mirror imaging systems based on few-shot machine learning

Area of Science:

• Optical Engineering
• Machine Learning Applications

Background:

• Off-axis reflective imaging systems present assembly challenges.
• Current design methods for easy assembly lack generalizability.

Purpose of the Study:

• To propose an automated generation method for easy-assembly off-axis three-mirror imaging systems.
• To develop a generalizable design process overcoming limitations of existing methods.

Main Methods:

• Constructing a novel few-shot dataset of optical system parameters meeting assembly and design constraints.
• Training a support vector regression (SVR) model using this dataset for automated parameter generation.
• Refining predicted mirror surface parameters with the Wassermann-Wolf (W-W) method to create freeform surfaces.

Main Results:

• The SVR model rapidly and reliably generates parameter combinations for off-axis three-mirror imaging systems.
• The method successfully integrates design requirements with assembly constraints.
• Freeform surfaces are created from predicted mirror parameters.

Conclusions:

• The proposed automated method provides a straightforward approach for designing integrated off-axis three-mirror imaging systems.
• This technique enhances the generalizability of easy-assembly design processes.
• The study demonstrates effective application of few-shot learning principles in optical system design.