摘要
为改善全变差正则化变量分离与增广拉格朗日(SUnSAL-TV)算法求解的丰度存在过平滑与边界模糊的现象,本文提出结构张量全变差(STV)再优化的稀疏解混算法(SUnSAL-TV-STV),用STV正则项校正SUnSAL-TV算法求解的丰度矩阵。本文在合成数据集与真实高光谱数据集上进行算法仿真,合成数据实验结果表明:本文算法与其他算法相比,解混重建误差提高0.01~0.03且具有最高的解混成功率,通过对真实数据解混丰度图的观察,本文算法较好地修复了SUnSAL-TV算法求解丰度图的过平滑与边界模糊现象。
The total variation-regularized variable separation and abundance values obtained by the augmented Lagrange(SUnSAL-TV)algorithm are too smooth and the boundary is blurred.To address this issue,in this paper,we propose a sparse unmixing algorithm known as structural-tensor total-variation(STV)re-optimization(SUnSAL-TV-STV)sparse unmixing.The proposed algorithm uses an STV regularizer to correct the abundance matrix obtained by the SUnSAL-TV algorithm.We performed algorithm simulations on synthetic datasets and real hyperspectral datasets,and the synthetic-data experiment results show that the proposed algorithm improves the unmixing reconstruction error by 0.01~0.03,and when compared with other algorithms,it achieves the highest unmixing success rate.The unmixed abundance maps of the real data indicate that the proposed method corrects the over-smoothing and boundary-blur issues that characterize the abundance maps obtained by the SUnSAL-TV algorithm.
作者
崔颖
王恒
朱海峰
CUI Ying;WANG Heng;ZHU Haifeng(College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001, China)
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2020年第7期1087-1093,共7页
Journal of Harbin Engineering University
基金
国家自然科学基金项目(61675051)
教育部博士点基金项目(20132304110007)。
关键词
高光谱遥感
光谱解混
稀疏解混
空间信息
结构张量
全变差
重建误差
解混成功率
hyperspectral remote sensing
spectral unmixing
sparse unmixing
spatial information
structure tensor
total variation
reconstruction error
unmixing success rate
