Deep learning-based single-shot lateral shearing interferometry

Abstract

Lateral shearing interferometry (LSI) is a powerful measurement method for wavefront sensing and optical testing. However, traditional LSI methods often face challenges in terms of complicated system configurations and vibration sensitivity. In this paper, we propose a novel approach that leverages deep learning to enable single-shot LSI for surface measurement. In our LSI system, the x- and y-directional shearing modules are attached together and a polarization grating and a polarization camera are utilized to obtain a single composite interferogram, which is the summation of the x- and y-directional shearing interferograms. Deep learning is then employed to accurately obtain the x- and y-phases (which are directly related to the surface slope) from the single composite interferogram, significantly reducing the effect of vibration and improving the robustness of the measurements. We trained a deep learning network using training data obtained from a deformable mirror so that the trained network knows how to retrieve the x- and y-phases from a single composite interferogram. We demonstrate the effectiveness of our approach through experimental measurement of different surfaces ranging from simple concave to complex random surfaces, and show that our deep learning-based LSI enables single-shot and even dynamic surface measurement. This work opens new avenues for the application of artificial intelligence in LSI to enable high-speed and dynamic measurement of specular surfaces.