Unified Full‐Stokes Polarimetry and Orbital Angular Momentum Metrology Leveraging Isotropic D 2 NN

ABSTRACT Polarization and phase constitute the complete vector‐wavefront information of structured light, which is essential in optical metrology and communication. However, single‐shot acquisition of vector wavefront remains constrained by complex algorithmic iterations and demanding device fabrication. Here, we propose a versatile isotropic deep diffractive neural network (D 2 NN) framework to transcend these limitations. By leveraging the D 2 NN, we demonstrate the ability to decode a longitudinally polarized field related to spin–orbit coupling for full‐Stokes parameters, while simultaneously reconstructing the longitudinal wavefront vortex phase. Our results show that, using a single phase‐only mask, the D 2 NN‐based detection geometry achieves an ultrahigh accuracy (> 99%) in full‐Stokes retrieval and an extremely low error (< 0.5%) in orbital angular momentum (OAM) spectrum reconstruction, coupled with remarkable noise robustness. More impressively, preliminary experimental diffraction field patterns mediated by longitudinal fields are in agreement with D 2 NN‐enabled theoretical results, enabling unified full‐Stokes polarimetry and OAM metrology. Compared with prior vector wavefront reconstruction approaches, the isotropic D 2 NN eliminates iterative computation and explicit polarization control by exploiting light‐speed forward inference. The proposed D 2 NN offers a promising pathway for industrial snapshot joint metrology requiring simultaneous polarization and phase information.