基于复值特征子空间的高光谱图像去噪  

Hyperspectral Image Denoising Based on Complex-Valued Feature Subspace

作　　者：牟奇春[1] MOU Qi-chun(School of Software,Chengdu Polytechnic,Chengdu 610041,China)

机构地区：[1]成都职业技术学院软件学院,成都610041

出　　处：《西南师范大学学报（自然科学版）》2020年第10期89-96,共8页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：四川省教育厅2019年重点项目(18ZA0170)。

摘　　要：针对传统的高光谱图像去噪方法忽视高光谱图像强烈的谱间相关性和图谱合一的问题,提出了一种基于复值特征子空间的高光谱图像去噪方法.该方法首先基于光谱数据的相似性对复域信号进行奇异值分解(Singular Value Decomposition,SVD)分析,选择最能代表信号子空间的最优低维特征子空间,然后基于非局部复域块匹配3D(Complex-Domain Block-Matching 3D,CDBM3D)滤波器对特征图像进行滤波.实验结果表明,本文算法对噪声具有较强的鲁棒性,可以有效恢复低信噪比的高光谱数据.与其他方法相比,本文方法在所有波长的RRMSE值最小的情况下准确性最佳.To solve the problem that,in the traditional denoising method of hyperspectral image,the strong spectral correlation and spectral integration of hyperspectral image have been ignored,a new denoising method based on complex value feature subspace has been proposed.According to this method,singular value decomposition(SVD)analysis has first been performed on complex-valued signals,and the optimal low-dimensional feature subspace that best represents the signal subspace been selected.And then the feature image has been filtered based on the non-local complex domain block matching 3D(CDBM3D)filter.The experimental results show that the algorithm has strong robustness to noise and can effectively recover hyperspectral data with low SNR.Compared with other methods,the method in this paper has the best accuracy with the lowest RRMSE value of all wavelengths.

关 键 词：高光谱图像 奇异值分解 噪声滤波 特征子空间 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]