基于全变分正则项的CASSI数据重构算法

A Reconstruction Algorithm of CASSI Data Based on Total Variation Regular Terms

在编码孔径快照光谱仪(Coded Aperture Snapshot Spectral Imager,CASSI)光谱重构算法设计中,两步迭代收缩阈值(Two-Step Iterative Shrinkage/Threshold,TwIST)算法实现了CASSI的光谱重构,但因正则项选取和引入的噪声随迭代次数而不断放大,导致其光谱重构精度低于80%。文章在TwIST算法基础上,以光谱图像具有空间平滑过渡特性为先验知识,提出两点改进:一是选择全变分正则约束项;二是对每一步迭代的更新项进行全变分去噪处理。为了验证改进后的算法,文章通过计算机仿真CASSI的采集数据,得出仿真数据光谱重构精度为90.93%;并根据CASSI样机采集试验数据,得出试验数据的光谱重构精度为86.56%。改进后的算法可为以后CASSI数据重构提供参考。

In designing spectral reconstruction algorithm for Coded Aperture Snapshot Spectral Imager(CASSI),the two-step iterative shrinkage/threshold(TwIST)algorithm can realize spectral reconstruction of CASSI,but the accuracy of spectral reconstruction is lower than 80%due to selection of regular terms and amplification of the introduced noise with the increase of iteration number.In this paper,based on the study of the traditional TwIST algorithm,two improvements are proposed,selecting the total variation regular constraint terms and denoising the updated terms in each iteration.In order to verify the improved algorithm,the data of CASSI are simulated by computer,with the spectral reconstruction accuracy 90.93%.The CASSI prototype is built to collect the experimental data,and the spectral reconstruction accuracy of the experimental data is 86.56%.The research results provide a reference for the following CASSI data reconstruction.