基于自适应加权的超光谱图像分类方法

Adaptive weight based hyperspectral image classification

为了进一步提高已有的加权超光谱图像分类方法的识别性能,以支持向量机(SVM)作为基本框架,结合超光谱图像的谱间与空域信息,提出了一种自适应加权的超光谱图像分类方法。所提方法的自适应权值是由两部分构成,即改进的具有归一化取值的互信息(MI)与一定量的归一化图像标准方差之和,不仅考虑到了超光谱图像的谱间信息对分类会产生的作用,同时也考虑到了每个波段图像所含信息对分类产生的作用,为基于加权的超光谱图像分类方法提供了一种新的思路。实验结果表明,本文提出的方法是有效的和可行的。

In order to improve the identification ability of the existing weight method,based on support vector machines（SVMs） for hyperspectral image classification,an adaptive weight method is proposed.The proposed method is made up of two parts,namely,the improved mutual information with normalized value and the normalized image standard deviation quantified by an adjustable coefficient.This method considers the effects on the classification from not only the spectrum of hyperspectral image but also the information contained in each band image.Therefore a new idea is provided for the weight based hyperspectral image classification method.The experimental results show that the proposed method is effective and feasible.