Research On Image Reconstruction Algorithm For Coded Aperture Spectral Imaging System

The spectral imaging technology acquires two-dimensional spatial information and one dimensional spectral information of the target scene,and constitutes a three-dimensional data cube.The data cube can provide a lot of useful information,so the spectral imaging technology has important application value in many fields.Traditional spectral imaging technology has some problems,such as low luminous flux and spending much time in scanning.The amount of raw data collected is usually large,which puts great pressure on data storage and transmission.The coded aperture spectral imaging technology is a new type of computational spectral imaging method based on compressed sensing theory.It encodes and modulates the spatial information and spectral information of the target scene through a coded template and a dispersed element to obtain two-dimensional compressed measurement values,and then uses an optimized algorithm to reconstruct the target spectral image.This technology breaks through the limitation of Nyquist sampling theorem very well,which has attracted widespread attention from scholars.In this paper,the coded aperture spectral imaging technology is studied in depth.Because the measured values obtained by the imaging system are highly compressed,the reconstruction of spectral images becomes a serious ill-conditioned inverse problem,and the research on reconstruction algorithms becomes an important challenge.Therefore,this paper focuses on the spectral image reconstruction algorithm in the coded aperture spectral imaging system.The main works as follow:Firstly,the basic theory of compressive sensing is introduced.Three main problems of this theory are explained in detail include the sparse representation of the signal,the construction of the measurement matrix and the signal reconstruction algorithm.At the same time,the basic principles of the coded aperture spectral imaging technology are introduced.And then the composite structure and mathematical model of the dual disperser design for coded aperture snapshot spectral imaging system and the single disperser design for coded aperture snapshot spectral imaging system are analyzed.Secondly,in view of the problem that traditional algorithms are not effective in reconstructing spectral images,an adaptive Split-Bregman iterative image reconstruction algorithm is proposed.This algorithm combines the Split-Bregman algorithm with dictionary learning,and constantly updates the estimated value of the target image and the redundant dictionary during the iterative solution process.Several simulation experiments based on single disperser coded aperture snapshot spectral imagers shows that the quality of the algorithm to reconstruct spectral images has obvious advantages when compared with traditional algorithms.Both the average peak signal-to-noise ratio and the average structural similarity are significantly improved.Finally,this article proposes an image reconstruction algorithm based on deep learning.The algorithm preprocesses the compressed measurement values of the imaging system firstly,and then establishes a deep convolutional network between the preprocessed results and the original image,finally obtains a non-linear mapping relationship through training and learning.Using the trained model can reconstruct the target spectral image.Several simulation experiments based on dual disperser coded aperture snapshot spectral imagers shows that this algorithm can not only improve the reconstruction quality of the spectral image when compared with traditional algorithms and the adaptive Split-Bregman iterative algorithm,but also greatly reduce the time taken for the reconstruction process.