Evolved design for telecentric optical systems based on particle swarm optimization and semi-supervised learning

Abstract

Finite-distance systems are used extensively in the field of machine vision, and engineers need to rely on experience and mathematical basis to design such systems. In this paper, an evolved design method for finite-distance optical systems (telecentric systems) based on particle swarm optimization and semi-supervised learning is proposed. The method establishes a function value based on the Seidel aberration equation and innovatively introduces the Head-To-Tail method to generate the high-quality initial dataset in bulk. The feasibility of the method is verified with an example, and the user only needs to determine the magnification, working distance and NA to obtain a telecentric optical system directly. In this way, the time cost of searching for the initial structure and continuous trying is greatly reduced, and the design threshold is lowered. Meanwhile, it has a positive effect on shortening the design cycle of the whole lens and leaving a surplus for processing.