文档介绍：IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 6, JUNE 2005 1351
Kernel-Based Methods for Hyperspectral
Image Classification
Gustavo Camps-Valls, Member, IEEE, and Lorenzo Bruzzone, Senior Member, IEEE
Abstract—This paper presents the framework of kernel-based scenario-dependent, and sometimes needs a priori knowledge.
methods in the context of hyperspectral image classification, For these reasons, a desirable property of hyperspectral data
illustrating from a general viewpoint the main characteristics classifiers should be to produce accurate land-cover maps
of different kernel-based approaches and analyzing their prop-
erties in the hyperspectral domain. In particular, we assess when working with high number of features, low-sized training
performance of regularized radial basis function works datasets, and in presence of different noise sources [2].
(Reg-RBFNN), standard support vector machines (SVMs), kernel In the remote sensing literature, many supervised methods
Fisher discriminant (KFD) analysis, and regularized AdaBoost have been developed to tackle the multi- and hyperspectral data
(Reg-AB). The novelty of this work consists in: 1) introducing classification problem. A esful approach to multispectral
Reg-RBFNN and Reg-AB for hyperspectral image classification;
2) comparing kernel-based methods by taking into account the data classification is based on the use of artificial -
peculiarities of hyperspectral images; and 3) clarifying their theo- works [3] (., multilayer perceptrons (MLP) [4]–[7], radial
retical relationships. To these purposes, we focus on the accuracy basis function works (RBFNNs) [8]–[10]). However,
of methods when working in noisy environments, high input these approaches are not effective when dealing with a high
dimension, and limited training sets. In addition, some other im- number of spectral bands, since they are highly sensitive to
portant issues are discussed, such as the sparsity of the solutio