Shift-rotation absolute measurement method for irregular aperture optical surfaces based on deep learning

Abstract

Absolute measurement technology has received considerable attention in the field of optical metrology owing to its high accuracy. However, to ensure accuracy of measurement, the traditional method requires the use of expensive high-precision stages and considerable time for the precise adjustment of the measured surface. A shift-rotation absolute measurement method for irregular aperture optical surfaces based on deep learning is presented here. In this method, the measured surface needs to be tested in the original position and two randomly changed positions. Furthermore, a simple and effective shift-rotation prediction convolutional neural network that can learn the mapping relationships between the binarized images of the changed position and shift-rotation is designed. The shape of the measured surface with an irregularly shaped aperture is obtained by fitting orthogonalized Zernike polynomials. Compared with the current absolute measurement method for irregular apertures, this method is highly efficient and cost-effective for adjusting the measured surface. The simulation and experimental results demonstrate the accuracy and validity of this approach.