摘要
针对目前高光谱图像基于流形学习的无监督特征提取算法中只能够单独描述高维数据空间局部或者全局的几何结构,并且没有一种算法能够同时保持高维数据全局和局部的几何结构的问题,提出了一种基于全局和局部流形结构的无监督特征提取算法(GLMS)对高光谱图像进行特征提取.算法基于流形学习基本理论,需要建立两种保持流形结构的近邻图,分别用来描述数据的全局和局部的流形结构,通过求解广义特征值问题获得重构权值矩阵进而得到低维嵌入空间的最优投影,以达到降维的目的.在AVIRIS高光谱图像以及Indian Pine和Salina数据集上进行仿真对比实验,结果表明,提出的算法在分类精度和计算效率上有较好的提高.
High spectrum image manifold learning unsupervised feature extraction algorithm can only separate description based on high dimensional data space whether it is local or global geometric structure or not,there is no algorithm can also maintain a high dimensional data of global and local geometric structure.A novel unsupervised feature extraction algorithm is proposed based on global and local manifold structure（GLMS）for feature extraction of hyperspectral image.Manifold learning algorithms are based on the basic theory,which establishes two maintain manifold structure neighbor graphs,and it is applied to describe the data of the global and local manifold structure,by solving the generalized eigenvalue problem to obtain the optimal projection reconstruction weight matrix and then to get the low dimensional embedding space to achieve the purpose of reducing dimension.In the AVIRIS hyperspectral image tests on Indian Pine and Salina data set,experimental results show that the proposed algorithm has better improvement in classification accuracy and computational efficiency.
出处
《沈阳大学学报（自然科学版）》
CAS
2015年第4期283-288,共6页
Journal of Shenyang University：Natural Science
关键词
高光谱图像
无监督特征提取
全局和局部流形结构
流形学习
hyperspectral image
unsupervised feature extraction
global and local manifold
manifold learning
