Study On Pattern Recognition Method Of Orbital Angular Momentum In Ocean Turbulent Environment

In underwater wireless optical communication,the vortex beam carrying Orbital Angular Momentum(OAM)has the spatial spiral phase distribution property,which provides a new information modulation dimension resource for underwater wireless optical communication and can greatly improve the channel capacity and spectral efficiency.Therefore,The OAM pattern recognition methods have attracted extensive attention.The traditional recognition methods are interferometer method and diffraction method,but because the vortex beam is easily distorted by the perturbed spiral phase of ocean turbulence,which greatly affects the OAM pattern recognition of traditional methods,machine learning can extract information from the distorted vortex beam phase map,which has the advantages of high recognition accuracy and small complexity.In this thesis,Laguerre-Gaussian(LG)beam is selected as the research object,and the effects of different network models on the OAM pattern recognition under different temperature and salinity turbulence-causing ratioω environments are discussed,and the results are given based on Convolutional Neural Networks(CNN)model,Histogram of Oriented Gradient-Support Vector Machine(HOGSVM)model and the joint CNN-SVM model for OAM pattern recognition are given respectively.The main work of this paper specifically includes the following.(1)The basic principle of orbital angular momentum beam is explained,the generation of LG beam in OAM beam and its mathematical expressions are introduced,and the phase distribution and light intensity distribution of single-mode the LG beam and dual-mode LG beam are simulated using numerical values.(2)A random phase screen model of ocean turbulence is constructed,based on the basic principle of the power spectrum inversion method to generate the phase screen of ocean turbulence,and then the construction of the ocean turbulence channel is carried out to analyze the light intensity distribution and phase distribution of the LG beam and the distribution of the spiral spectrum of the LG beam under different ω ocean turbulence environments.(3)A CNN model is given,which uses its convolutional structure and automatic information extraction capability to use the LG beam phase maps after ocean turbulence as the data set,and the CNN model can effectively extract information from the distorted the LG beam phase maps to realize the recognition of the OAM single mode and the OAM double mode under different ω ocean turbulence.(4)The HOG-SVM model is given,and the HOG method is used to extract features from the spatial phase diagram of the LG beam,and the HOG features of the phase diagram are implemented to identify the OAM modes under different ω ocean turbulence environments.(5)A joint CNN-SVM model is given to study the recognition effect of the joint CNNSVM model in different ω ocean turbulence environments and to verify the feasibility of the model to achieve the OAM pattern recognition.Comparing three different machine learning models,the OAM single-mode recognition based on the joint CNN-SVM model is better than the CNN model,and the maximum difference between the OAM dual-mode recognition accuracy of the CNN-SVM model and the CNN model is 2.223%.the HOG-SVM model has the best classification recognition effect,and in the environment at ω=-1.75,The accuracy of the OAM((?)=1～10)singlemode recognition is as high as 93.778%,and the accuracy of the OAM((?)=±1～ ±10)dualmode recognition is as high as 100.000%,It can be concluded that the HOG feature method of phase graph extraction is more suitable for spiral phase wavefront characteristics of phase graph.