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

Integrated processing is an effective approach to reduce the assembly difficulty of off-axis reflective imaging systems. However, existing design methods for the easy assembly of off-axis reflective systems generally face specific design requirements, resulting in varying design processes and insufficient generalizability. This study proposes an automated generation method for easy-assembly off-axis three-mirror imaging systems, utilizing a support vector regression (SVR) model inspired by few-shot machine learning principles. First, a novel approach, to our knowledge, to construct a few-shot dataset where all parameters of off-axis three-mirror optical imaging systems meet both assembly constraints and design requirements simultaneously is proposed to serve as the foundation for training the SVR model. Then, an SVR model designed to automatically generate parameter combinations for off-axis three-mirror spherical imaging systems is built and trained using the constructed dataset, thus facilitating the design process. Finally, based on design requirements and assembly constraints, the SVR model predicts suitable parameter combinations for the three-mirror imaging systems, and the predicted mirror surface parameters are further refined using the improved Wassermann-Wolf (W-W) method to create freeform surfaces. The experimental results demonstrate that the method presented in this study achieves rapid and reliable attainment of the off-axis three-mirror imaging system that satisfies both design and assembly criteria, providing a straightforward approach for designing the integrated off-axis three-mirror imaging system.