Deep Learning Ghost Polarimetry of Two-Dimensional Objects with Amplitude Anisotropy

The paper discusses the possibilities of deep learning in solving the inverse problem of computational ghost polarimetry. For the first time it is shown that the spatial distribution of the polarization properties of objects with linear amplitude anisotropy is restored using a neural network trained on model data. The spatial distribution of the parameters of linear amplitude anisotropy is determined with an accuracy of 7.8 and 15.6% for azimuth of anisotropy and value of anisotropy, respectively.