Evaluation Method for Band Selection Algorithm of Hyperspectral Image

As a dimensionality reduction method,band selection is a significant preprocessing step used for classification applications of hyperspectral image data.And a large number of algorithms have been proposed for band selection.However,the evaluation method for band selection algorithms was rarely studied.In order to evaluate the performance of band selection algorithms,an evaluation method is proposed.The method is implemented by measuring the selected band sets from three aspects,i.e.,information content,class separability and correlation.In experiment,the evaluation method is applied to four common band selections algorithms.The results of experiments show that the evaluation method is feasible in practical applications.