Optical design of off-axis three-mirror reflective system by neural networks

Abstract

We incorporate neural networks into the optical design of off-axis three-mirror reflective system, enabling us to achieve design outcomes without relying on iteration or ray tracing methods. Our approach involves combining analytical relations with neural networks during the design process, which yields results covering the entire parameter space with a single user input, and each design is scored simultaneously. Our results demonstrate that neural networks can simulate the complex relationship between performance requirements and structural parameters of an optical system. As such, the structural parameters can be directly obtained from the performance requirements, replacing the iterative optimization process traditionally used. This approach leads to relatively efficient and straightforward optical design. We anticipate that this method can be extended to various optical systems, reducing the experience threshold and difficulty of optical design.