Band selection algorithm based on information entropy for hyperspectral image classification

A band selection algorithm based on information entropy is proposed for hyperspectral image classification. First, original spectral features are transformed into discrete features and represented by a discrete space model. Then, the band selection algorithm based on information entropy is adopted to reduce feature dimensionality. The bands with weak class separability are effectively abandoned by the band selection algorithm. Moreover, support vector machine classifiers with composite kernels are employed to incorporate spatial features into spectral features, reducing speckle errors in the classification maps. The proposed methods are applied to three benchmark hyperspectral data sets for classification. The performance of the proposed methods is compared with a band selection algorithm based on mutual information. The experimental results demonstrate that the band selection algorithm based on information entropy can effectively reduce feature dimensionality and improve classification accuracy.