Deep-learning-based recognition of fractional C -point indices in polarization singularities

Abstract

Polarization singularities are ubiquitous in most of the vector beams and can be divided into different morphologies as characterized by the polarization singularity index. A precise identification of different polarization topologies may provide abundant polarization morphology for singular optics and increase the information-carrying capacity with a vector beam. However, discrimination of the fractional polarization singularity indices is still an untapped study area. Here, based on deep learning of the Stokes polarimetry, an efficient method is proposed to accurately identify the fractional $C$-point indices in polarization singularity for the vector light fields. Various vector light fields carrying different fractional $C$-point indices are prepared by a polarization-controlled Sagnac interferometer with an embedded spatial light modulator. Then, by analyzing the measured Stokes parameters with a training set of the residual convolutional neural network, the identification of a fractional polarization singularity index with an accuracy up to 0.01 is achieved. The method applies to all $C$-point degenerate beams and may hold promise for increasing the information channel capacity with vector light.