Robust occlusion-aware orbital angular momentum feature extraction via all-optical diffractive processing systems.

Area of Science:

• Optical Computing
• Artificial Intelligence
• Photonics

Background:

• Diffractive optical neural networks (DONNs) offer high-speed, energy-efficient optical information processing.
• Conventional DONNs lack robustness against occlusions and distortions, limiting their real-world applicability.
• Orbital Angular Momentum (OAM) offers unique properties for optical information encoding.

Purpose of the Study:

• To develop a robust all-optical feature extraction framework for DONNs using OAM.
• To enhance the resilience of optical computing architectures to occlusions and environmental uncertainties.
• To enable stable and efficient information representation in the OAM domain for improved DONN performance.

Main Methods:

• Proposed a novel feature extraction framework utilizing Orbital Angular Momentum (OAM).
• Employed a deep learning-trained diffractive processing framework to convert optical information into OAM modes.
• Investigated the self-adaptive, occlusion-immune capabilities of the OAM-based DONN.

Main Results:

• The OAM-based DONN framework demonstrated high performance across diverse and irregular occlusions.
• The system maintained accuracy without requiring additional network retraining or samples.
• Achieved stable and efficient information representation in the OAM domain.

Conclusions:

• The OAM-based approach significantly enhances the robustness and scalability of DONNs.
• This method enables reliable optical computing in dynamic, occlusion-rich environments.
• Paves the way for next-generation optical computing systems with improved real-time processing capabilities.