Convolutional-neural-network-assisted parameter identification in elliptical Airy vortex beams.

Area of Science:

• Optics and Photonics
• Machine Learning Applications
• Optical Communications

Background:

• Convolutional neural networks (CNNs) are powerful tools for image recognition.
• Elliptical Airy vortex beams (EAVBs) offer expanded dimensionality for orbital angular momentum (OAM) communication.
• Classifying EAVB parameters is crucial for leveraging their communication potential.

Purpose of the Study:

• To develop a compact CNN architecture for classifying the topological charge (m) and elliptical parameter (t) of EAVBs.
• To train and validate the CNN using a physics-augmented dataset combining simulated and experimental data.
• To assess the network's robustness on unseen experimental EAVB patterns.

Main Methods:

• Development of a compact CNN architecture tailored for EAVB parameter classification.
• Training the CNN on a dataset augmented with physics-informed simulations and experimental intensity patterns.
• Testing the network's accuracy and generalization capabilities on standard and unseen experimental datasets.

Main Results:

• The CNN achieved over 99.80% accuracy in classifying EAVB parameters (m and t) on a standard test set.
• The network demonstrated high robustness on unseen experimental patterns, with accuracy ranging from 98.24% to 100%.
• All unaugmented new experimental patterns were recognized accurately by the CNN.

Conclusions:

• The developed compact CNN is highly effective for classifying key parameters of EAVBs.
• This machine learning approach paves the way for EAVB-based optical communications with enhanced capacity and dimensionality.
• The findings support the integration of AI in optical systems to meet increasing bandwidth demands.