基于边信息的高光谱图像恢复模型  

Hyperspectral image restoration model with side information

作　　者：张少杰 罗琼[1,2,3] 韩志[1,2] 唐延东[1,2] Zhang Shaojie;Luo Qiong;Han Zhi;Tang Yandong(State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,China;Institutes for Robotics&Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang 110016,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院沈阳自动化研究所机器人学国家重点实验室,沈阳110016 [2]中国科学院机器人与智能制造创新研究院,沈阳110016 [3]中国科学院大学,北京100049

出　　处：《计算机应用研究》2021年第10期3166-3171,3195,共7页Application Research of Computers

基　　金：国家自然科学基金资助项目(61903358);国家自然科学基金面上项目(61773367);国家自然科学基金创新群体项目(61821005);中国科学院青年创新促进会资助项目(2016183)。

摘　　要：在高光谱图像(HSI)恢复中,如何在模型中有效嵌入先验信息和正确建模噪声一直是研究的两个重点。边信息作为一种基于域的先验知识已经在许多方向取得了成功,然而在高光谱去噪领域仍未受到关注。为了将这种领域知识与高光谱恢复模型自然耦合,提出的方法采用双线性映射的方式将边信息链接到表示观测数据潜在低秩结构的底层矩阵,并使用E-3DTV(enhanced 3-D total variation)正则编码了HSI局部平滑先验。此外该方法使用L p范数进行噪声建模,进一步增强对腐败的鲁棒性。该方法在两个数据集、七种加噪方式下与五种竞争方法在三个数值指标上进行了比较,结果充分反映了提出方法对复杂噪声场景的有效性和鲁棒性。For hyperspectral image(HSI)restoration,how to effectively embed prior information in the model and correctly model the noise have always been the two focus of research.As a domain-dependent prior knowledge,side information has succeeded in many aspects,but it has not received much attention in the field of hyperspectral denoising.In order to naturally couple this domain knowledge with the hyperspectral restoration model,the method linked side information to the underlying matrix representing the potential low-rank structure of the observation data via a bilinear mapping,and used E-3DTV to encode HSI local smoothness prior.In addition,this method used the L p norm for noise modeling to further enhance the robustness against corruption.This method was compared with five competitive methods on three numerical indicators in two data sets and seven noise addition methods.The results fully reflect the effectiveness and universality of the proposed method for complex noise scene.

关 键 词：边信息 低秩矩阵学习 高光谱图像去噪 L p范数 增强三维全变分 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]