Recognition of Two-Mode Optical Vortex Beams Superpositions Using Convolution Neural Networks

We investigate the efficiency of convolutional neural networks (CNNs) application for recognition of two-mode optical vortex (OV) beams superpositions. Unlike standard multiplexing, we associate information channels not with individual modes, but with pairs of modes with a given index difference which raises security of information transmission. At the first stage, we performed studies with a model dataset using standard image augmentation techniques for training CNNs (translation and rotation). Further, we use experimentally generated by phase spatial light modulator (SLM) intensity patterns for training the proposed neural networks. The achieved test accuracy of the CNNs trained on the experimentally generated dataset is 0.84. This value is comparable with the test accuracy for the modeling training dataset.