Accurate polarization prediction of metasurfaces via physics-constrained deep learning
Metasurface-enabled polarization manipulation has attracted broad interest for compact polarimetric imaging and detection. However, full-wave electromagnetic simulations are time-consuming, and direct polarization state prediction is difficult. Here, we develop a physics-constrained deep learning framework for rapid prediction and reconstruction of metasurface polarization responses over 650–1000 nm. Building on a constrained sine/cosine phase representation, the model jointly predicts the transmittance and phase of two orthogonal polarized electric fields, with unit-circle normalization enforcing valid phase components and a sigmoid mapping bounding the transmittance outputs. The dual-channel responses are converted into polarization states and evaluated on the Poincaré sphere. The model achieves a phase mean absolute error of approximately 0.02 rad and provides an efficient approach for polarization-response prediction.
Authors
- Bobo Du (ORCID: https://orcid.org/0000-0003-2507-7613)
- Yanpeng Zhang (ORCID: https://orcid.org/0000-0003-3827-8983)
- Yihan Chen
- Siqi Liu
- Lei Zhang (ORCID: https://orcid.org/0000-0002-5113-1786)
- Yuqi Zeng
- Jingping Zhu (ORCID: https://orcid.org/0000-0002-6226-2247)
Publication Details
- Journal
- Optics Letters
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1364/ol.612037
- Primary Topic
- Metamaterials and Metasurfaces Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China
- Natural Science Basic Research Program of Shaanxi Province