Instant reconstruction of the longitudinal component in tightly focused fields using polarimetric data and deep neural networks

This work presents a data-driven approach for reconstructing the longitudinal component of tightly focused optical fields using only experimentally accessible polarimetric intensity images. A custom-designed deep neural network is trained on simulated polarimetric mappings generated from aberrated wavefronts through a high-NA objective. The model successfully reconstructs the complex amplitude of the longitudinal field with high fidelity, offering a practical and instant method for indirect measurement of longitudinal components in tightly focused beams.