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

Convolutional neural networks (CNNs), widely employed in image recognition, show significant promise in identifying topological charges of vortex beams. Elliptical Airy vortex beams (EAVBs) introduce an additional elliptical parameter <i>t</i>, expand the dimensionality of Airy vortex beams, and thus offer an opportunity to increase the capacity for orbital angular momentum communication. In this work, we develop a compact CNN architecture to classify key parameters of EAVBs: the topological charge <i>m</i> and the elliptical parameter <i>t</i>. Trained on physics-augmented datasets that combine simulated and experimental autofocusing intensity patterns, the network achieves over 99.80% accuracy on a standard test set across three independent training runs. Its robustness is further validated on a set of unseen experimental patterns: it accurately recognizes all unaugmented new patterns and maintains consistent performance (98.24%-100%) on augmented variations. These results lay the groundwork for EAVB-based optical communications enhanced by machine learning, providing large capacity and high dimensionality to meet growing bandwidth requirements.