基于类别可分性的高光谱图像波段选择

Supervised Band Selection Based on Class Separability for Hyperspectral Image

高光谱遥感数据具有丰富的光谱信息,应用十分广泛,但其冗余的光谱信息有时会限制高光谱图像的分类等的精度以及计算复杂度。为了提高解译效率,高光谱图像降维不可或缺,这也是高光谱图像处理的研究热点之一。提出了一种基于类别可分性的高光谱图像波段选择方法(Endmember Separability Based band Selection,ESBB),该方法通过Mahalanobis距离最大化图像中各类地物的可分性来确定最优的波段组合。相较于其他监督波段选择算法,该方法不需要大量训练样本,不用对每个组合做分类处理。对波段选择后的结果进行分类的实验结果证明,该方法是一个快速有效的波段选择方法,可以得到一个较好的分类精度。

Hyperspectral remote sensing data contain abundant spectral information,which are widely used,but sometimes the redundant spectral information limits the classification accuracy and computational complexity.In order to improve the interpretation efficiency,hyperspectral image dimension reduction is necessary,which is also one of the highlights in the hyperspectral image processing.This paper presented a hyperspectral image band selection method based on endmember separability（EBSS）.This method maximizes class separability using Mahalanobis distance to determine the optimal band combination.Compared with other supervision of band selection algorithms,the proposed method does not need the training samples,and does not conduct classification during band selection.The experiment results show that the proposed method is effective and can get better classification accuracy.