基于正交指数局部保留投影的高光谱图像特征提取

A hyperspectral image feature extraction method based on orthogonal exponential locality preserving projections

针对高光谱图像,在判别局部保留投影(Discriminant Locality Preserving Projection,DLPP)的基础上,提出了一种名为正交指数判别局部保留投影(Orthogonal Exponential Discriminant Locality Preserving Projection,OEDLPP)的特征提取方法。该算法不但保留了DLPP算法的有监督特性,还利用了指数矩阵(the matrix exponential)来获取更有效的样本信息,避免了小样本问题。同时,OEDLPP对投影矩阵进行施密特正交化,解决了特征的冗余性问题。应用OEDLPP算法对高光谱图像进行特征提取后,并采用支持向量机(SVM)对降维后的数据进行分类。与主成分分析(PCA)、局部保留投影(LPP)、判别局部保留投影(DLPP)、指数判别局部保留投影(EDLPP)、正交判别局部保留投影(ODLPP)等对比实验结果表明,本文算法对样本有效信息的获取具有一定的优越性,分类精度提升了2%~3%左右。

For hyperspectral images,we proposed a feature extraction method called orthogonal exponential discriminant locality preserving projection(OEDLPP) based on discriminant locality preserving projection(DLPP).The OEDLPP algorithms not only retains the supervised characteristic of the DLPP algorithms,but also utilizes the matrix exponential to obtain more effective sample information,which can avoid the Small Samples Size problem.Meanwhile,Schimidt orthogonalization is used to the projection matrix in OEDLPP,which solved the problems of feature redundancy.Last but not least,we used support vector machine(SVM) to classify the hyperspectral images after using OEDLPP algorithms to extract feature.Compared with several existing algorithms,such as principal component analysis(PCA),locality preserving projection(LPP),discriminant locality preserving projection(DLPP),exponential discriminant locality preserving projection(EDLPP) and orthogonal discriminant locality preserving projection(ODLPP),the proposed algorithms has a certain superiority for obtaining the effective information of the sample,and the classification accuracy is improved by about 2%~3%.