Deep-learning-based recognition of multi-singularity structured light.

Area of Science:

• Optics and Photonics
• Quantum Information Science
• Artificial Intelligence

Background:

• Structured light with tailored topological patterns is crucial for classical and quantum research.
• Current detection methods struggle with complex multi-singularity phase structures, limiting analysis.
• Accurate recognition of general structured light with multiple singularities remains a significant challenge.

Purpose of the Study:

• To develop a novel deep learning (DL) framework for end-to-end unveiling of multi-singularity phase structures.
• To enable direct phase output from intensity patterns for detailed analysis of twisted photons.
• To propose a new phase-based optical secret sharing (OSS) protocol utilizing multi-singularity modes.

Main Methods:

• A deep learning framework was trained using only two intensity patterns of structured light.
• The DL model directly outputs phase information, bypassing limitations of traditional methods.
• The framework was validated for Laguerre-Gaussian (LG) modes and general phase objects.

Main Results:

• The DL framework successfully identifies complex multi-singularity phase structures in an end-to-end manner.
• The system accurately acquires phase information for various structured light modes, including LG modes.
• A novel OSS protocol was demonstrated using multi-singularity modes, offering enhanced security and state space.

Conclusions:

• The developed DL platform provides an accurate and versatile tool for analyzing complex structured light.
• This technology facilitates the creation of advanced optical secret sharing protocols.
• The findings open new possibilities for applications in high-capacity communications, laser analysis, and microscopy.