张量环分解和空谱全变分的高光谱图像恢复算法  

Hyperspectral Image Restoration Algorithm Based on Tensor Ring Decomposition and Spatial-spectral Total Variation

作　　者：陈千 罗显康[1] 谢巧玉 李霞 CHEN Qian;LUO Xiankang;XIE Qiaoyu;LI Xia(Faculty of Science,Yibin University,Yibin,Sichuan 644000,China)

机构地区：[1]宜宾学院理学部,四川宜宾644000

出　　处：《宜宾学院学报》2024年第12期21-28,共8页Journal of Yibin University

基　　金：宜宾学院培育项目(2022PY30)。

摘　　要：高光谱图像在采集和转换中会受到各种污染,目前在Tucker或CP上进行的多数去除噪声算法会改变信号固有的结构,对张量秩的最优估计非常困难.为此,提出基于张量环分解和空间光谱全变分的高光谱图像恢复模型:利用张量环分解的张量核范数和空间光谱全变分来约束低秩,更好地探索全局空间结构和相邻波段的频谱相关性,并利用增广拉格朗日算法求解此模型.数值实验表明,模型去除噪声后的图像清晰,PSNR、SSIM和FSIM三个指标均优于现有算法.Hyperspectral images were subject to various pollutants during acquisition and conversion,and most noise removal algorithms currently used on Tucker or CP could alter the inherent structure of the signal,making it very difficult to estimate the optimal tensor rank.To this end,a hyperspectral image restoration model based on tensor ring decomposition tensor and spatialspectral total variation was proposed:low rank was constrained by tensor ring decomposition tensor kernel norm and spatialspectral total variation to better explore the global spatial structure and spectral correlation of adjacent bands,and the augmented Lagrangian algorithm was used to solve this model.Numerical experiments show that the model produces clear images after removing noise,and the PSNR,SSIM,and FSIM metrics are all superior to existing algorithms.

关 键 词：张量核范数 高光谱图像 张量环分解 空间光谱全变分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]