A wavelet-based band selection method for hyperspectral image classification

This paper presents an adaptive wavelet-based band selection method for hyperspectral image classification, that simultaneously selects relevant bands by analysing few spectral signatures of classes of interest. The properties of the wavelet transform in representing signal non-stationarity enable the definition of an adaptive and direct procedure where the number of bands is not pre-defined. Preliminary experimental results show that the method is able to identify informative bands for the input spectra, reaching classification accuracies that often are comparable to those obtained using the full data. In addition, it adapts to the specific characteristics of the input data, while being computationally efficient.