Research On Image Reconstruction Method Of Snapshot Multispectral Imaging Syste

Spectral imaging technology can obtain two-dimensional spatial information and one-dimensional spectral information of an object.Spatial information can be used to determine the size and other external characteristics of an object.Spectral information can be used to analyze the internal characteristics of objects such as composition and content.The spectral characteristics of spectral imaging technology make it widely used in many fields.Traditional spectral imaging usually requires scanning to obtain spatial and spectral 3D data,the data acquisition time is long,and the hardware system usually has scanning moving parts.As a new type of spectral imaging implementation,snapshot spectral imaging only collects one image through one exposure,and can obtain two-dimensional spatial information and one-dimensional spectral information of an object by combining reconstruction algorithms.This paper mainly studies coded aperture snapshot spectral imaging,using digital micromirrors for random coding,to study and compare different reconstruction algorithms.Firstly,the principle of coded snapshot spectral imaging and compressive sensing reconstruction algorithms are introduced,including Gradient Projection for Sparse Reconstruction(GPSR)and Alternating Direction Method of Multipliers(ADMM).Secondly,in order to further improve the quality of reconstructed spectral images,a spectral image reconstruction algorithm based on neural network is proposed and introduced.Then,the public data set is used to simulate the test chart of the single grating coding spectral imaging method,and the simulation analysis of the multi-spectral reconstruction is carried out for the three algorithms of GPSR,ADMM and neural network.Although the neural network algorithm has a long training time,the reconstruction time is relatively shorter,and the reconstruction of spectral images is better.Finally,a single grating snapshot spectral imaging system based on a digital micromirror was initially designed and constructed,and a neural network algorithm was used to reconstruct the laser image collected by the system.The experimental results show that the laser spot images of the two wavelengths can be accurately separated,and the laser spectral position is more accurate.