Band selection in spectral imaging for classification and regression tasks using information theoretic measures

In this paper we present three different methodologies of band selection for hyperspectral data sets applied to classification and regression tasks using Information Theory measures. In one of the cases, the bands will be selected having information about the classification labels of the data points (supervised classification). In the second one, no information about the target labels is required (unsupervised classification). In the third problem, the target variables are of continuous nature and are also available (supervised regression).