Deep-learning-based colorimetric polarization-angle detection with metasurfaces

Polarization plays a key role in both optics and photonics. Generally, the polarization states of light are measured with birefringent or dichroic optical elements paired with a power meter. Here we propose a direct polarization detection method based on colorimetric asymmetrical all-dielectric metasurfaces to obtain the polarization angles of the incident light. The independently tunable periods and diameters along the x and y axes enables double-layer nanopillars to realize high-performance dual-color palettes with arbitrary combinations under orthogonal polarization states. The polarization detection network based on residual networks is used to deeply learn the regulations between color palette variations and incident polarization angles, which can accurately recognize extremely slight polarization variations in about 1 s with an accuracy of 81.4% within 0.7 degrees error and 99.5% within 1.4 degrees error. Our strategy significantly improves the compactness of polarization detection, and it can be readily expanded to polarization distribution measurement and colorimetric polarization imaging on an intelligent platform. (C) 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement

Automated summary

This research proposes a direct polarization detection method based on colorimetric asymmetrical all-dielectric metasurfaces, which can accurately measure the polarization angles of light. The double-layer nanopillars with independently tunable periods and diameters enable high-performance dual-color palettes, and the polarization detection network based on deep learning can accurately recognize slight polarization variations.