Broadband single-shot full-Stokes polarization detection enabled by metasurface vector holography and a neural network algorithm

Polarimetry plays important roles in fields such as optical sensing, imaging, and communication. Current polarization detection schemes based on metasurfaces provide a feasible way for developing compact polarimeters. However, existing approaches face challenges that include insufficient accuracy, limited bandwidth, and complex operation/analysis procedures. Here, we propose a generic polarimetric scheme based on metasurface vector holography, capable of highly accurate, broadband, robust, and single-shot full-Stokes detection of arbitrary polarizations. This method encodes the polarization information (amplitude and phase of pairs of left- and right-hand circularly polarized light) of normally incident light into the intensity distribution of vector holographic images and then deciphers the polarization and intensity of light from the images by a trained neural network model. We show that the proposed scheme can achieve accurate polarization detection covering the entire Poincaré sphere in a broad wavelength range of 630-1200 nm. The average reconstruction errors of azimuthal and ellipticity angle at a wavelength of 635 nm can reach 0.58272° and 0.6715°, respectively, which is comparable to that of commercial polarimeters. Furthermore, due to the nature of holographic encoding and neural network decoding strategy, the method is immune to metasurface fabrication errors. Our design features high detection accuracy, wide operating wavelength range, robustness to fabrication errors, and direct readout from a single measurement. This study provides an efficient scheme for compact full-Stokes polarimetry and shows potential applications in fields including miniaturized optical systems, real-time sensing, and polarization imaging.