基于低秩和稀疏模型的高光谱图像快速去噪方法  

Fast denoising method of hyperspectral image based on low rank and sparse model

作　　者：杨垚 黄聪 王华军[1,2] YANG Yao;HUANG Cong;WANG Huajun(Chengdu University of Technology Department of Geophysical Institute,610059,China;Chengdu University of Technology Department of Key Laboratory of Earth Exploration and Information Technology of Ministry of Education,610059,China)

机构地区：[1]成都理工大学地球物理学院,成都610059 [2]成都理工大学地球勘探与信息技术教育部重点实验室,成都610059

出　　处：《物探化探计算技术》2021年第5期663-668,共6页Computing Techniques For Geophysical and Geochemical Exploration

基　　金：四川省人工智能重点实验室项目(2020RYJ02)。

摘　　要：随着高光谱遥感影像(HSI)研究热点的不断上升,影像的去噪工作越显重要。利用HSI影像的特殊特征(频谱间强相关性和低秩子空间等)和混合噪声的特性,提出了基于低秩稀疏矩阵分解的迭代算法,用以除去多种类型的混合噪声。这里提出的算法SLRMS(Subspace Low-Rank Matrix and Sparse Matrix Factorization)充分利用HSI频谱低秩特性,在低秩和稀疏正则化的约束下迭代达到去噪的效果。提出的算法在模拟小数据集Indian pines和大数据集KSC(Kennedy Space Center)上去噪后的视觉效果和定量评价指标,均表现优越且运行所费时间极低。With the continuous rising of research hotspots on hyperspectral remote sensing images(HSI),denoising of images becomes more and more important.The use of the special characteristics of HSI images(strong correlation between spectra and low-rank subspace,etc.)and the characteristics of mixed noise,an iterative algorithm based on low-rank sparse matrix decomposition,is proposed to remove multiple types of mixed noise.The algorithm SLRMS(Subspace Low-Rank Matrix And Sparse Matrix Factorization)proposed in this study makes full use of the low-rank characteristics of the HSI spectrum,and iteratively achieves the denoising effect under the constraints of low-rank and sparse regularization.The algorithm has excellent visual effects and quantitative evaluation indicators after denoising on the simulated small data set Indian pines and the large data set KSC(Kennedy Space Center),and the its running time is extremely low.

关 键 词：高光谱遥感影像 去噪 混合噪声 频谱高相关性 低秩子空间 

分 类 号：P631.4[天文地球—地质矿产勘探]