基于线性嵌入和张量流形的高光谱特征提取  

Feature Extraction Based on Linear Embedding and Tensor Manifold for Hyperspectral Image

作　　者：马世欣 刘春桐 李洪才 张耿[2] 何祯鑫 Ma Shixin;Liu Chuntong;Li Hongcai;Zhang Geng;He Zhenxin(College of Missile Engineering, Rocket Force University of Engineering, Xi'an , Shaanxi 710025, China;Key Laboratory of Spectral Imaging Technology, Xi'an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi' an, Shaanxi 710119, China)

机构地区：[1]火箭军工程大学导弹工程学院,陕西西安710025 [2]中国科学院西安光学精密机械研究所光谱成像技术重点实验室,陕西西安710119

出　　处：《光学学报》2019年第4期127-135,共9页Acta Optica Sinica

基　　金：国家自然科学基金(41574008;61501465)

摘　　要：为了使降维结果更好地体现高光谱数据的空间结构信息,并进一步提高分类精度,提出了一种基于线性嵌入和张量流形的高光谱特征提取算法。不同于其他流形结构的表达方法,所提算法采用协同表示理论求解全局线性嵌入的权重矩阵,更有利于保持高维数据的全局信息,提高了流形结构表达的准确性。同时,建立了基于多特征描述的张量流形降维框架,得到的显式映射具有较强的可靠性和全局适应性。实验结果表明:与主成分分析、局部线性嵌入、拉普拉斯特征映射和线性保留投影等算法相比,所提算法表现出了更优越的分类性能。In order to express the spatial structure information of hyperspectral image more effectively and improve the classification accuracy after dimensionality reduction, we propose a hyperspectral feature extraction algorithm based on linear embedding and tensor manifold. Different from other manifold structure expression methods, the proposed algorithm uses the cooperative representation theory to solve the weight matrix for globally linear embedding, which is more beneficial to maintain the global information of high dimensional data and improve the accuracy of manifold structure expression. At the same time, the dimension reduction framework of tensor manifold based on multi-feature description is established, and the obtained explicit mapping has strong reliability and global adaptability. Experimental results show that compared with the principal component analysis, locally linear embedding, Laplacian Eigenmap, linearity preserving projection and other algorithms, the proposed algorithm has better classification performance.

关 键 词：遥感 高光谱 降维 线性嵌入 流形学习 张量表达 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]