基于张量环分解的非局部高光谱图像去噪算法  

Nonlocal regularized hyperspectral image denoising algorithm based on tensor ring decomposition

作　　者：陈千 陈利霞[1,2] CHEN Qian;CHEN Lixia(School of Mathematics and Computing Science,Guilin University of Electronic Technology,Guilin 541004,China;Guangxi College and Universities Key Laboratory of Data Analysis and Computation,Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学数学与计算科学学院,广西桂林541004 [2]桂林电子科技大学广西高校数据分析与计算重点实验室,广西桂林541004

出　　处：《桂林电子科技大学学报》2021年第2期146-153,共8页Journal of Guilin University of Electronic Technology

基　　金：国家自然科学基金(11961010,61941111);广西自然科学基金(2018GXNSFAA138169)。

摘　　要：高光谱图像去除噪声的2个问题:1)在采集和运输的过程中会产生各种各样的噪声,使得人们无法快速且准确地获得信息;2)大部分去除噪声算法都在Tucker或CANDECOMP/PARAFAC(CP)上进行,而Tucker或CP是将高维信号转化为低维,改变了信号固有的结构,对于张量秩的最优估计很难,且涉及的参数使得计算量很大。针对以上问题,提出一种基于张量环分解的非局部正则化高光谱图像去除噪声算法(TRTD-NRM)。该算法利用张量环分解直接处理高维信号的特质来研究全局光谱相关性(GCS)和空间非局部自相似性(NSS),能够易于计算和很好地保留高光谱图像的内在性质。通过设计一种交替方向乘子法来求解模型。数值实验结果表明,此方法得到的去除噪声后的图像很清晰。与现有算法相比,本算法无论是从主观效果还是客观评价上均具有较强的可比性。Hyperspectral images have rich spectral features and widely application.There are two problems in removing noise from hyperspectral images,on the one hand,in the acquisition and transmission process,hyperspectral images(HSIs)is unavoidably corrupted by several types of noise which makes people unable to obtain information quickly and accurately.On the other hand,most of the traditional denoising algorithm are carried out on Tucker or CANDECOMP/PARAFAC(CP),which converts high-dimensional signals into low-dimensional signals,thus changing the inherent structure of the signals.Inherent structure,it is difficult to optimally estimate the rank of tensor,and the parameters involved make the amount of calculation large.To solve the above problems,this paper proposes a non-local regularized hyperspectral image denoising(TRTD-NRM)based on tensor ring decomposition.The algorithm uses the characteristics of Tensor ring decomposition algorithm directly processes high-dimensional signals to study global spectral correlation(GCS)and spatial non-local self-similarity(NSS),which can be easily calculated and retain the inherent properties of hyperspectral images.Design an alternating direction multiplier method to solve the model.The image after removing the noise is very clear.Numerical experiments show that the noise-removed image obtained by this method is very clear.Compared to existing algorithms,the proposed algorithm is highly competitive both in subjective and objective effects.

关 键 词：图像去噪 高光谱图像 张量环分解 潜在张量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]