An Off-Axis Flight Vision Display System Design Using Machine Learning

Abstract

We propose an off-axis flight vision display system design with a free-form surface using machine learning to simulate the visual distance variation during take-off and landing training for pilots. This design is realized by ray tracing using ZEMAX software, where we build and optimize a series of initial systems that meet the corresponding optical specifications. A deep neural network is used to train the regression model, which is specifically designed to predict the fitted polynomial model for the free-form surface of the system. Our results demonstrate that the design of a flight visual display system can be transformed into a machine learning problem and further optimized by training and learning with abundant data, providing an avenue to design more powerful and complex imaging optical systems.

Design of a Miniaturized Wide-Angle Fisheye Lens Based on Deep Learning and Optimization Techniques

“…We applied a deep learning (DL) algorithm to process the lens data in the wide-angle fisheye lens design. Deep learning is a subfield of machine learning based on artificial neural networks [24][25][26][27]. The first challenge of applying DL to optical design is that a deep neural network (DNN) model can be trained to predict the response of a given design, but the opposite is not possible.…”