Neural network assisted high-spatial-resolution polarimetry with non-interleaved chiral metasurfaces

Abstract

Polarimetry plays an indispensable role in modern optics. Nevertheless, the current strategies generally suffer from bulky system volume or spatial multiplexing scheme, resulting in limited performances when dealing with inhomogeneous polarizations. Here, we propose a non-interleaved, interferometric method to analyze the polarizations based on a tri-channel chiral metasurface. A deep convolutional neural network is also incorporated to enable fast, robust and accurate polarimetry. Spatially uniform and non-uniform polarizations are both measured through the metasurface experimentally. Distinction between two semblable glasses are also demonstrated. Our strategy features the merits of compactness and high spatial resolution, and would inspire more intriguing design for detecting and sensing.