Research On Measurement-Driven Framework For Ghost Imaging Method

Ghost imaging is a novel imaging technique that uses the secondorder intensity correlation of the computed optical field to obtain the image of a target object.This unique imaging method divides the optical path into two parts,where the signal path collects object information by a bucket detector without spatial resolution,and the reference path records the optical field information via a detector with spatial resolution.The target object image can be reconstructed by using correlation algorithm to calculate the information from both paths.This computation-based imaging method has excellent anti-interference capability,so it has great potential in various fields.However,ghost imaging often requires a large amount of samples to reconstruct the image of the target object,resulting in slower imaging speed.The large amount of data transmission also increases the requirements of hardware and makes it difficult to deploy in practical scenarios.Therefore,this paper focuses on exploring how to improve the imaging performance of ghost imaging at low sampling rates.And,the main research contents are as follows:Firstly,the speckle patterns of ghost imaging is optimized by the measurement-driven framework theory to improve the imaging quality of ghost imaging at low sampling rates.In this paper,we present a new speckle patterns optimization scheme based on the measurement-driven framework with a low-rank constraint on speckle patterns to decrease coherence between speckle patterns.Simulation results show that compared with other speckle patterns optimization schemes,this scheme performs better in terms of peak signal-to-noise ratio(PSNR)and structural similarity index(SSIM)at low sampling rates.Secondly,deep learning theory is used to remove the strong background noise in ghost imaging.After optimizing the speckle patterns via compressive sensing theory,the gray-scale speckle patterns is obtained.If a large-sized object needs to be reconstructed,the gray-scale speckle patterns will become inconvenient due to hardware limitations.Therefore,it is necessary to convert the gray-scale speckle patterns into a binary scattering pattern.However,after binarization,a large amount of background noise appears and becomes more and more obvious as the sampling frequency increases.To address this problem,the paper employs a deep learning denoising network to process imaging results with noise.The simulation results indicate that using this approach can significantly remove background noise.