高光谱遥感影像降维的拉普拉斯特征映射方法

Laplacian Eigenmap for Hyperspectral Remote Sensing Image Dimensionality Reduction

针对高光谱遥感影像数据量大、数据冗余度高的特点,引入拉普拉斯特征映射方法对高光谱遥感数据进行非线性降维。为了解决传统流形学习方法不能处理大数据量遥感影像的问题,本文提出了基于多元线性回归的拉普拉斯特征映射线性解法。实验证明,本文提出的降维方法能够保持数据集在原始特征空间分布的局部几何属性,降维后的影像具有更好的分类精度。

Feature extraction is an indispensable preprocessing step for large and high redundancy data of hyperspectral remote sensing image.In this paper,a Laplacian Eigenmap（LE）is introduced for dimensionality reduction.In order to overcome the shortcoming of conventional manifold learning which could not deal with large data,a linearization procedure for LE is proposed based on multiple linear regression analysis.The experiment demonstrates that the proposed dimensionality reduction method can preserve local geometry of samples in original feature space,the low dimensionality image could achieve a better classification accuracy.