Computational Ghost Imaging: A Study Based On Neutron Sources And Deep Learning Algorithms

Ghost imaging(GI),an unconventional imaging method that utilizes the intensity correlation of the light field to recover the information of the object,has received increased attention over the past few decades because of its advantages of high sensitivity,high resolution and low costs.This dissertation presents the related research works and background knowledge during the Ph D study and are organized as follow.The first part is the introduction.We mainly introduce the basic principle of ghost imaging,so as to get the advantages of ghost imaging compared with other imaging methods;then we briefly introduced the research progress of ghost imaging technology,and analyzed the bottlenecks existing in the current research.The significance and importance of this research are discussed in two aspects: neutron ghost imaging and high-speed ghost imaging.The second section present the neutron energy spectrum computational ghost imaging.We simply introduce the physical basis of neutron imaging,neutron sources and detectors,traditional neutron imaging technology,and the bottleneck of neutron imaging technology,then we discuss the breakthroughs that neutron ghost imaging can make in the field of neutron imaging.We also describe the neutron ghost imaging experiments in detail.We propose a new method to fabricate high resolution thermal neutron masks,and select a gas bucket detector suitable for time of flight according to the pulse characteristics of spallation source,thus we get high spatial resolution ghost imaging of neutron energy spectrum for the first time.Compared with traditional neutron spectral imaging,neutron ghost imaging is easy to carry out,which can achieve high spatial resolution and high temporal resolution at the same time,and is table-top and low cost.The third section is the simulation and experiments of high-speed ghost imaging based on a heuristic algorithm and deep learning.We introduce the basic principle of deep learning,commonly used networks,and we combine the theory of ghost imaging,make a new interpretation for ghost imaging from the angle of the convolutional neural network,which explains the deep learning algorithm can greatly improve the ghost image contrast and signal-to-noise ratio.Then,we introduced the principle of highspeed ghost imaging.We report an overlapping sampling scheme to accelerate computational ghost imaging for imaging moving targets for the first time,and we use a heuristic algorithm to get the coding matrix to decrease redundant information.In addition,we used deep learning algorithm and specially designed a network structure suitable for overlapping sampling scheme.Detailed comparisons show that we can greatly shorten the imaging time without losing the image information,so as to improve its application value.Finally,we summarize all the work of this dissertation,and look forward to the development of relevant research fields and the development of the follow-up work.