基于光谱梯度角的高光谱影像流形学习降维法

发布时间：2018-04-14 14:35

本文选题：光谱学 + 流形学习　；参考：《激光技术》2017年06期

【摘要】：为了挖掘高光谱数据的光谱局部特征,从高光谱遥感数据内在的非线性结构出发,提出了一种基于光谱梯度角的高光谱影像流形学习降维方法。采用局部化流形学习算法局部保持投影(LPP)对高光谱遥感数据进行非线性降维,对距离度量进行改进,将能够更好刻画高光谱影像光谱局部特征的光谱梯度角相似性度量应用于LPP方法,并用真实高光谱图像进行降维实验,取得了优于LPP方法和采用光谱角的LPP方法的结果。结果表明,在光谱规范化特征值方面,所提方法优于LPP方法和采用光谱角的LPP方法;在信息量的保持方面,具有更好的局部细节信息保持量。采用光谱梯度角的流形学习方法用于高光谱影像降维能取得较好的降维效果。
[Abstract]:Based on the intrinsic nonlinear structure of hyperspectral remote sensing data, a dimensionality reduction method for hyperspectral image manifold learning based on spectral gradient angle is proposed in order to mine the local spectral characteristics of hyperspectral data.The local manifold learning algorithm (LPP) is used to reduce the nonlinear dimension of hyperspectral remote sensing data, and the distance measurement is improved.The similarity measure of spectral gradient angle which can better depict the local characteristics of hyperspectral images is applied to LPP method. The dimension reduction experiments of real hyperspectral images are carried out. The results obtained are better than that of LPP method and LPP method with spectral angle.The results show that the proposed method is superior to the LPP method and the LPP method based on spectral angle in terms of spectral normalized eigenvalues, and has better local detail information retention in terms of information retention.Using the manifold learning method of spectral gradient angle to reduce the dimension of hyperspectral image, a better dimensionality reduction effect is obtained.
【作者单位】：空军航空大学航空航天情报系;
【基金】：吉林省教育厅“十二五”科研项目(2015448) 吉林省科技发展计划资助项目(20140101213JC)
【分类号】：TP751

【相似文献】