基于波段选择的高光谱遥感影像分类

Band Selection Based Hyperspectral Remote Sensing Image Classification

针对高光谱数据波段众多、数据量较大的特点,提出了一种基于波段选择的高光谱遥感影像分类方法,以北京昌平小汤山地区高光谱遥感数据为例,分析了各波段的信息含量和相邻波段的相关性,采用子空间划分、自适应波段选择的方法,实现了特征波段的选择。针对农村道路和空地、柏油路和居民地间的同谱异物现象,利用J-M距离模型判别其类间的可分性,获得了最佳波段组合,最后采用支持向量机分类器进行分类。结果表明,采用波段选择的方法能有效地提高高光谱数据的分类精度。

We put forward a hperspectral remote sensing image classification method based on band selection.Taking hyperspectral data（PHI sensor getting 80 band） data in Beijing Changping Xiaotangshan area for example,we analyze the content of information and the adjacent band correlation in each band.Then we use subspace division and adaptive band selection method to realize the feature band selection.Experimental results show that,the band selection method can effectively improve the classification precision of hyperspectral data.