摘要
针对传统稀疏非负矩阵分解(NMF)解混方法仅考虑丰度矩阵中非零个数最少,没有考虑混合像元内端元的丰度分布具有不均匀性的这一问题,提出一种基于信息熵的NMF遥感图像解混算法.将端元的丰度值的大小看成是信息熵中的符号出现的概率,当端元等概率出现在混合像元中时各个丰度值大小相等,对应的实际地物等比例出现在混合像元中,此时信息熵最大,但是丰度稀疏性最低;当丰度分布最不均匀时,仅有一种地物类型出现,信息熵最小,此时丰度值的稀疏性最高,只有一个非零值,由此得出丰度稀疏性和信息熵有负相关的关系.在NMF解混算法的基础上,引入负信息熵来约束丰度矩阵,同时加入平滑限制来约束端元光谱矩阵.在模拟数据和真实数据上进行了结果测试.实验结果表明:相比传统的NMF解混算法和基于l2范数的NMF遥感图像解混算法,本方法能得到更好的解混效果.
Aiming at the problem that the traditional sparse non-negative matrix factorization(NMF) unmixing algorithm only considers the minimum number of non-zero abundances while neglects the non-uniformity of endmember abundances,an information entropy-based remote sensing image unmixing algorithm for NMF was proposed.The values of abundances were regarded as the probability of symbols appearing in the information entropy.When the endmember appeared in uniform probability in the mixed pixels,the magnitude of each abundance value was equal,and the corresponding proportion of the actual objects appeared in the mixed pixel.At this time,the information entropy was the largest,but the abundance sparsity was the lowest.When the distribution was the most inhomogeneity,only one type of land feature appeared,and the information entropy was the lowest.At this time,the sparsity of abundance value was the highest,and there was only one non-zero value.Thus,relationship between sparsity and information entropy was obtained.Based on the NMF unmixing algorithm,the negative information entropy was used to constrain the abundance matrix,and another smoothing restriction was used to constrain the endmember spectral matrix.The results of synthetic data and real data were tested.Experimental results show that the proposed method has better performances than traditional NMF unmixing algorithm and l2 norm-based NMF unmixing algorithm.
作者
李杏梅
王伟奇
LI Xingmei;WANG Weiqi(College of Mechanical and Electronic Information,China University of Geosciences(Wuhan),Wuhan 430074,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2019年第11期25-29,共5页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(61771437)
关键词
非负矩阵分解
遥感图像解混
信息熵
最少非零个数
不均匀性
non-negative matrix factorization(NMF)
remote sensing image unmixing
information entropy
least non-zero number
inhomogeneity
