摘要
为了解决高光谱遥感影像的特征融合问题,针对高光谱数据的维数高、信息量繁杂冗余、非线性而且数据量庞大特点,利用图谱理论非负稀疏保持嵌入的降维方法,提出基于光谱空间结合的非负稀疏保持嵌入的谱聚类进行样本的标记算法,有效地利用空间信息和原有光谱信息,提高分类的精度。该算法在引入非负稀疏表示的同时,利用样本的光谱与空间相关信息构建Laplacian图,嵌入投影到低维的子空间,然后再用经典的K均值聚类算法进行分类。算法能够有效保持样本的几何稀疏结构,而且光谱空间信息的结合使得图像的边界像素点得到了更好的分类。
To solve the characteristic fusion of hyperspectral remote sensing images, i. e. , using the .dimensional reduction method of non-negative sparse preserving embedding, this paper proposed spectral clustering algorithm based on the hybrid of spectral and spatial information and non-negative sparse preserving embedding for efficiently improve the classification accuracy with the spatial information and the original spectral information. This algorithm embedded the projection into low-dimensional subspace. After that it classified with the K-means clustering algorithm. The proposed algorithm can effectively maintain the geometry sparse structure of samples and make the boundary pixels of the image have excellent classification using the hybrid of the spectral and spatial information.
出处
《计算机应用研究》
CSCD
北大核心
2015年第6期1917-1920,共4页
Application Research of Computers
基金
航空科学基金资助项目(201210P8003)
关键词
非负稀疏
降维
谱聚类
高光谱图像
拉普拉斯
non-negative sparse
dimensionality reduction
spectral clustering
hyperspectral remote sensing image
Laplacian
