摘要
针对高光谱遥感数据特征提取方法的研究,提出了一种新的监督近邻重构分析(Supervised Neighbor Reconstruction Analysis,SNRA)算法。该方法首先利用同一类别的近邻数据点对各数据点进行重构;然后在低维嵌入空间中保持该重构关系不变,尽可能地分离开非同类数据点,并利用总体散度矩阵来约束数据间的相关性;最后求解得到一个最佳投影矩阵,进而提取出鉴别特征。SNRA算法不仅保持了同类数据的局部结构而且增强了非同类数据的可分性,同时减少了数据的冗余信息。在Indian Pine和KSC高光谱遥感数据集上的实验结果表明:提出的方法能更好地揭示出高光谱遥感数据的内在特性,提取出更有效的鉴别特征,改善分类效果。
For the feature extraction methods of hyperspectral remote sensing data, a new method, called supervised neighbor reconstruction analysis （SNRA）, was proposed. First, this method reconstructs each point with neighbor points from the same class. Then, it preserves the reconstruction relationship and separates the data points from different classes as far as possible in low-dimension embedding space. And a total scatter matrix is used to constrain the correlation between data points. Finally, it obtains an optimized projection matrix and extracts the discriminating feature. SNRA not only preserves the local structures of intraclass data but also enhances the separability of interclass data. And it reduces the redundant information. The experiments on Indian Pine and KSC hyperspectral remote data sets show that the proposed method can better reveal the intrinsic property of hyperspectral remote sensing data and effectively extract the discriminating feature to improve the classification result.
出处
《红外与激光工程》
EI
CSCD
北大核心
2016年第10期271-278,共8页
Infrared and Laser Engineering
基金
国家自然科学基金(理论物理专项)(11547196)
四川省教育厅重点资助(15ZA0229)
四川理工学院人才引进资助(2013RC07)
关键词
高光谱遥感数据分类
特征提取
监督学习
邻域重构
总体散度矩阵
hyperspectral remote sensing data classification
feature extraction
supervised learning
neighbor reconstruction
total scatter matrix
