Photon‐Starved Full‐Stokes Polarimetric LiDAR Imaging Using a Fractal SNSPD and a Learning‐Based Denoiser

ABSTRACT We implement a learning‐based denoiser to the full‐Stokes polarimetric imaging LiDAR based on a fractal superconducting nanowire single‐photon detector. The denoiser is a four‐branch denoising network that combines local and global feature fusion with polarimetric loss functions for optimal noise removal and physical accuracy. After training the denoiser with real‐life data, the denoiser processes the polarimetric images acquired with short pixel‐dwell time and outputs the images with high qualities previously achievable by increasing the pixel‐dwell time. Therefore, the denoiser mitigates the tradeoff between image quality and acquisition time, permitting full‐Stokes polarimetric LiDAR imaging in the extremely photon‐starved condition.