Research On Data Mining And Reconstruction Algorithm Optimization For High Quality Ghost Imaging

With the rapid development of intelligent medical treatment,unmanned driving,intelligent security and other fields,how to quickly obtain high-quality reconstructed images has become the goal of China’s intelligent construction and other related fields.However,the traditional "point-to-point" imaging method has poor imaging quality under low radiation and low light due to the limitations of the surface array detector’s inability to respond,which cannot meet the imaging requirements in intelligent fields.The emergence of ghost imaging technology,which separates the reconstruction process from the detection process,breaks through the response limit and detection space limitation of array detector,and has unique advantages such as strong anti-interference ability,long acting distance and all-weather monitoring,providing necessary means for target detection under low dose or limited light intensity environment.However,the physical quantity carrying the spatial characteristics of the target in the process of ghost imaging detection is scattered and hidden in a large number of detection data,which makes the ghost image is often buried by noise,and it is difficult to judge the spatial shape and texture details of the target.As a result,in ultra-low radiation medical imaging and stealth target detection,major medical and malignant events such as tumor or cancer cell characteristic diagnosis error and detection failure are likely to occur.Therefore,the study of high-quality ghost imaging can provide new ideas for target recognition and detection in harsh environments.In order to solve the above problems,this paper improves the quality of ghost imaging from two dimensions of ghost imaging detection data mining and reconstruction algorithm optimization.The main work of the research is as follows:(1)This paper studies the data mining strategy from the traditional point of view and proposes a ghost imaging optimization method based on double threshold data mining.Firstly,the effects of bucket detector sorting and speckle pattern sorting on ghost image reconstruction are analyzed through comparative experiments.Secondly,double thresholds are introduced to mine the sorted detection data so as to increase the difference between speckle pattern data.Finally,the feasibility and effectiveness of the proposed method in data mining and the universality of the mined data are verified by numerical simulation.In addition,three joint data mining models are also designed as new ideas of data mining,which lays a foundation for the practical application of ghost imaging.(2)A self-encoded neural network model is constructed from the perspective of deep learning,and a deep learning ghost imaging optimization method based on data mining is proposed.This method is optimized and improved based on the existing self-encoded denoising convolutional neural network to make it suitable for ghost imaging reconstruction.The useful features were mined through model training,and the results of rough imaging reconstruction were input to the trained network model for verification.The results show that the proposed method can clearly reconstruct the object under 0.1 sampling rate,and it takes less time to reconstruct a single image.The method in this chapter opens up a new way for super-resolution imaging and fast imaging and has important promotion value.(3)From the perspective of reconstruction algorithm,the method of highlighting the edge details of reconstructed image under low sampling rate was studied,and a ghost imaging reconstruction model based on ADMM-TV regularization was proposed.This method utilizes TV regularization and L2 norm to construct a reconstruction model,At the same time,this paper adopts alternating multiplier iterative algorithm as the model solving method,which can not only simplify the solving process but also improve the operation speed and increase the robustness.In addition,the paper finally verifies that the proposed method has better antinoise performance under noisy conditions,and the more complex the grayscale image,the better the anti-noise performance.