基于核Fisher判别分析的高光谱遥感影像分类

Hyperspectral Remote Sensing Image Classification Based on Kernel Fisher Discriminant Analysis

高光谱遥感技术,将反映目标辐射特性的光谱信息与反映目标空间位置关系的图像信息有机地结合在一起。高光谱影像具有丰富的光谱信息,较全色、多光谱影像能够更好的进行地面目标的分类识别。在介绍核Fisher判别分析算法的基础上,选用径向基核函数,使用一对一或一对余构造多类构造法,并利用交叉验证网格搜索法优化核函数参数,构建了快速稳定的多类核Fisher判别分析分类器。通过OMIS和AVIRIS影像的分类实验,表明了核Fisher判别分析与支持向量机的分类精度相当,但是所需的训练时间较短。

The hyperspectral remote sensing technology,which appeared early in 1980s,combines the radiation information which relates to the targets' attribute,and the space information which relates to the targets' position and shape,completing the information continuum of optics RS image from panchromatic image to hyperspectral via multi-spectral image.The spectrum information,which is rich in the hyperspectral image,compared with panchromatic remote sensing image and multispectral remote sensing image,can be used to classify the ground target better.It has become an important technique of map cartography,vegetation investigation,ocean remote sensing,agriculture remote sensing,atmosphere research,environment monitoring and military information acquiring.As Support Vector Machine(SVM) was applied to machine learning fields successfully in recent years,the classic linear pattern analysis algorithms which was called the 3rd revolution of pattern analysis algorithms,can cope with the nonlinear problem.Some references applied the kernel methods to linear Fisher Discriminant Analysis(FDA),and put forward Kernel Fisher Discriminant Analysis(KFDA).Firstly,this paper introduced the classification method based on the kernel fisher discriminant analysis.For the binary problem,the aim of FDA is to find out the linear projection(projection axes) on which the intra-class scatter matrices of the training samples are maximized and scatter matrices of inter-class are minimized.For KFDA,the inputted data is mapped into a high dimensional feature space by a nonlinear mapping,while linear FDA in the feature space will be performed.Secondly,we researched on the selection methods of the kernel function and its parameter,and studied on the multi-classes classification methods,and then applied them to hyperspectral remote sensing classification.We use decomposition methods of multi-class classifier and method of parameter selection using cross-validating grid search to build an effective and robust KFDA classifier.Finally,we carried out the hyperspectral image classification experiments based on KFDA and some other comparative experiments.Some conclusions can be drew as follows.Using the kernel mapping,the KFDA experiment on PHI and AVIRIS image demonstrates that the KFDA is less affected by the dimension of input sample,and can avoid the Hughes phenomena effectively.The results show that it has more comparable classification accuracy than support vector machine classifier.There is no need to compute the complicated quadratic optimizing problem in training KFDA classifier as SVM classifier does,so this algorithm is not very complicated and costs less time.Especially in the one-against-rest decomposition,comparing with the SVM,KFDA is much faster.The capability of KFDA classifier is affected a lot by kernel function and its parameters,and a fine recognition precision can only be obtained when the kernel function's parameters are appropriate.The stability of classification can be effectively improved by parameter selection via cross-validate grid search method.