Phase Recovery Of Fractional Vortex Beams Based On Deep Learning

Optical vortex beams are able tocarry quantized orbital angular momentum(OAM)and unlimited OAM states supply extra channels for data multiplexing,which provide one of the promising methods for improving the channel capacity of optical communication.Compared with integer-order vortex beams,the fractional-order vortex beams can provide higher-precision channel resolution.The phase information of a fractional optical vortex changes in the diffraction process,this allows the communication based on the fractional vortex optical mode to have more encoded information.Therefore,obtaining the phase information of fractional vortex beams at different diffraction positions is of great significance for expanding the application scenarios of vortex beam communication.Based on this background,the main research contents of this paper are as follows:1.A scheme based on generative adversarial networks was proposed to restore the phase of fractional order vortex optics.First,the intensity images of fractional vortex in the initial position,Fresnel diffraction zone and Fraunhofer diffraction zone are obtained through simulation.Then input the collected intensity images into the model for training.The model can recover the phase information of the beam at the corresponding position at the output end based on the intensity information of the beam at a certain diffraction position obtained from the input end.Compared with the previous deep learning model for fractional order vortex beams,this model can not only obtain the modes of fractional order vortex beams,but also obtain phase images of the beam at three diffraction positions and accurately recover the number of phase singularities.On the corresponding data set,the restoration accuracy of the initial position model is 100%,the Fresnel diffraction area is 98.4%,and the Fraunhofer diffraction area is 87.5%.The average recovery accuracy of the model is 95.3%.2.A three-dimensional OAMSK color image information restoration scheme based on fractional order vortex light was proposed.On the basis of the phase recovery model,the sender separates a color image into three channels in RGB format,and then selects 8fractional order vortex beam modes under 4 different phase changes.The intensity images at three different diffraction positions are used to encode the images in each format.At the receiving end,a phase recovery model is used to demodulate the beam intensity,and then the recovered beam pattern is pixel decoded to obtain the corresponding image information in the channel.Finally,the decoded three channel images are composite to restore the original color image.Compared to previous grayscale image restoration systems,this scheme can obtain more image information.3.A color image information restoration system using fractional vortex beam diffraction pattern modulation is proposed.At the sending end,a color image is separated into three channels in RGB format,and then the intensity images of the three diffraction positions of the fractional vortex beam are used for information modulation and pixel encoding.At the receiving end,the phase recovery model is used to demodulate the beam intensity,and then the recovered beam pattern is pixel-decoded to obtain the image information in the corresponding channel.Finally,the decoded three-channel images are composited to restore the original color image.3.Fractional order vortex optical phase recovery based on deep learning has been studied.Based on the mode distribution of the beam OAM spectrum,a dataset with 12 modes was constructed for residual phase vortex light.A dataset with 10 light field modes was constructed for multi fractional order vortex light.Using the previous phase recovery model for training and testing,the phase recovery accuracy reached 88.3% and 92.5% on the corresponding dataset,respectively.This is the first time that deep learning methods have been applied to the field of fractional order vortex optics.