Recognition of orbital-angular-momentum modes with different topological charges and their unknown superpositions via machine learning

Light beams carrying orbital-angular-momentum (OAM) play an important role in optical manipulation and communication owing to their unbounded state space. However, it is still challenging to efficiently discriminate different OAM modes and their superpositions. Here we demonstrate that neural networks can be trained to recognize OAM modes with different topological charges and unknown superpositions. Using intensity images of OAM modes generated in simulations and experiments as the input data, we illustrate that our neural network has the generalization power to recognize OAM modes beyond training areas with high accuracy, whose topological charges are up to $\ifmmode\pm\else\textpm\fi{}8$ and 104 for experimental and numerical data, respectively. Moreover, the trained neural network can correctly classify and predict arbitrary superpositions of two OAM modes with random topological charges. Our machine learning approach only requires a small portion of experimental samples and significantly reduces the cost in experiments.