Band Selection for Hyperspectral Image Classification using Mutual Information
Spectral band selection is a fundamental problemin hyperspectral data processing. In this paper, a new band-selection method based on mutual information (MI) is proposed.MI measures the statistical dependence between two randomvariables and can therefore be used to evaluate the relativeutility of each band to classification. A new strategy is describedto estimate the mutual information using a priori knowledgeof the scene, reducing reliance on a ‘ground truth’ referencemap, by retaining bands with high associated MI values (subjectto certain so-called ‘complementary’ conditions). Simulations ofclassification performance on 16 classes of vegetation from theAVIRIS 92AV3C dataset show the effectiveness of the method,which outperforms an MI-based method using the associatedreference map, an entropy-based method and a correlation-basedmethod. It is also competitive with the steepest ascent (SA)algorithm at much lower computational cost.