高光谱影像的近邻加权拉普拉斯降维方法
Dimensionality Reduction for Hyperspectral Images Based on Cam Weighted Distance Laplacian Eigenmap
路易 1郭静 2于少波1
作者信息
- 1. 装备学院 研究生管理大队, 北京 101416
- 2. 装备学院 复杂电子系统仿真实验室, 北京 101416
- 折叠
摘要
针对高光谱影像数据中存在信息冗余和非线性结构的现象,以及数据分布不均匀时拉普拉斯特征映射近邻点选择不恰当的问题,提出了一种基于Cam加权距离的拉普拉斯改进算法,用于高光谱影像数据降维以压缩数据量并提高分类精度.首先对波段分组去除奇异波段,然后用基于Cam加权距离的拉普拉斯特征映射算法对剩余数据降维,最后将结果输入最小距离分类器进行高光谱影像分类.通过Indiana Pines数据集进行验证,实验结果表明:与线性降维主成分分析法和非线性降维拉普拉斯特征映射相比,基于Cam加权距离的拉普拉斯特征映射算法分类精度更高.
Abstract
In consideration of the information redundancy and intrinsic nonlinearities, and the irrelevancy of Laplacian Eigenmap k-nearest neighbor selected for the uneven distribution of hyperspectral image data, this paper presents an improved LE algorithm based on Cam weighted distance for hyperspectral image dimensionality reduction to compact feature representation and improve the accuracy of classification.First, the band is grouped for the removal of singular band, then the Cam weighted distance Laplacian Eigenmap is used to reduce the remaining data dimension, and finally, the results are put into the minimum distance classifier for hyperspectral image classification.By verification with the Indiana Pines data set, the experimental results show that compared with linear dimensionality reduction method of PCA and nonlinear method of LE, Cam weighted distance Laplacian Eigenmap algorithm gets higher classification accuracy.
关键词
Cam加权距离/拉普拉斯特征映射/非线性降维/波段选择Key words
Cam weighted distance/Laplacian eigenmap (LE)/nonlinear dimensionality reduction/band selection引用本文复制引用
出版年
2017
NETL
NSTL
维普
万方数据