采用改进RPCA的遥感影像去云算法  

Cloud removal algorithm of remote sensing image via improved RPCA

作　　者：石晓旭 夏克文[1] 王宝珠[1] 常虹 武盼盼[1] SHI Xiao xu, XIA Ke-wen, WANG Bao-zhu, CHANG Hong, WU Pan-pan(School of Electronic and Information Engineering, Hebei University of Technology, Tianjin 300401, Chin)

机构地区：[1]河北工业大学电子与信息工程学院,天津300401

出　　处：《计算机工程与设计》2018年第6期1653-1658,共6页Computer Engineering and Design

基　　金：河北省自然科学基金项目(E2016202341);河北省高等学校科学技术研究基金项目(BJ2014013)

摘　　要：为解决遥感影像中常见的复杂稀疏云区的联合去云问题,提出一种采用改进的鲁棒主成分分析(RPCA)的遥感影像去云算法。根据遥感云层影像的自身特性,构造RPCA算法模型,采取一种新的基于分式函数的L0范数优化方式,引入加权核范数最小化算法(WNNM)对奇异值阈值进行自适应调节,提高云区矩阵的稀疏度和地貌矩阵的低秩性。实验结果表明,采用改进RPCA的遥感影像去云算法,能够去除复杂稀疏云区的云层遮挡,获得清晰度更高的无云遥感影像,在主观视觉和客观指标上均优于传统算法。To deal with the problem of common cloud removal in complex sparse cloud occlusion of remote sensing image,the improved robust principal component analysis（RPCA）algorithm to remove the cloud of remote sensing image was presented.Based on the property of remote sensing image,the RPCA algorithm was constructed and a new method based on fractional function to optimize the L0 norm was designed.Simultaneously,the weighted nuclear norm minimization algorithm（WNNM）was introduced to adjust the singular value threshold adaptively.Experimental results demonstrate that the proposed method can enhance the sparse property of the cloud matrix and the low-rank property of the landform matrix,remove the cloud in complex sparse cloud occlusion and get higher definition and more information remote sensing images without any cloud.It has better performance in subjective visual and objective indicator.

关 键 词：遥感图像去云 鲁棒主成分分析 加权核范数 分式函数 DC算法 自适应阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]