Recognition of orbital angular momentum vortex beam based on convolutional neural network

To identify different modes of orbital angular momentum (OAM) vortex beams after demultiplexing, deep learning technology is introduced, and a convolutional neural network (CNN) model is designed to detect OAM beams. The light intensity distribution maps of the Laguerre Gaussian beam with the topological charge from 1 to 20 through experiments are collected, the random phase screens are generated by using the power spectrum inversion method, to simulate the transmission of Laguerre Gaussian beams in the different atmospheric turbulence channels. The recognition accuracy is studied by the conditions of different CNN model iteration times, beam wavelengths, and data sets. The images are classified and processed to make different data sets, and the random vortex beam training sets are tested. Experimental results show that this method with about 98% accuracy for the conditions of long-wavelength beams and medium or weak turbulence. The training on complex training sets can improve the effect.