Lens Design Method Prediction of Local Optimization Algorithm by Using Deep Learning

Abstract

A design rule prediction is proposed to assist a lens design in this paper. Deep learning was applied in order to predict a lens design rule that is based on a local optimization algorithm. Three separate lens design rules related to the aperture stop and FOV variation were made for the optimization in the two-lens element optical systems whose structural parameters were created randomly. These random lens structures were optimized by using three separate lens design rules that were developed by Zemax OpticStudio API to create a big optimization dataset. All of the optimization results were collected by means of a further deep learning process to determine which optimization rule would be the better choice for lens optimization when given the lens parameters. The model developed via deep learning shows that the prediction has a 78.89% accuracy in determining an appropriate optimization rule for an assistant lens design.