Research On The Autoencoder And The Second-order Correlated Imaging

Second-order correlation imaging is a novel imaging method.The characteristics of two-way separation imaging and bucket detector make it have strong anti-turbulence ability,which has attracted wide attention.In recent years,in the field of computer vision,the algorithm based on machine learning has achieved great success in the fields of image classification,object detection,image segmentation and so on.At the same time,people began to introduce machine learning into the field of correlation imaging to reduce noise and improve image quality.However,most of the existing related work relies on computers,and the noise reduction ability of machine learning has not been fully studied.Based on the above problems,this paper studies the application of machine learning in second-order correlation imaging.The main work includes as follows:1.In this paper,the low-sampling second-order correlation imaging technology based on self-encoder is studied and the simulation experiment is carried out.At present,in the field of combining machine learning and second-order correlation imaging technology,the training and application of network model all need to rely on the computer,which restrict the application of the imaging technology.Based on this,this paper proposes the design scheme of Autoencoder-GI technology based on FPGA,and carries out the simulation experiment of Autoencoder-GI under the condition of low sampling rate(5.1%),and obtains a good imaging result,which verifies the feasibility of the scheme,and preliminarily shows that the actual system can run independently without relying on the computer.2.At the same time,this paper studies the influence of the noise in the bucket detector on imaging quality,in view of the situation,we proposed a network model(based on CNN Autoencoder-GI)training plan with better ability to resist the noise in the bucket detector,and based on the network model of the training scheme,we further studied its anti-noise effectiveness of multiplicative noise and additive noise.The simulation results show that adding noise to the hidden layer of the network model during training can effectively improve the ability of the imaging system to resist the noise of the bucket detector.