Unsupervised Band Selection by Integrating the Overall Accuracy and Redundancy

Abstract

Band selection is of great significance to alleviate the curse of dimensionality for hyperspectral (HSI) image application. In this letter, we propose a novel unsupervised band selection method for HSI classification. This method integrates both the overall accuracy and redundancy into the band selection process by formulating an optimization model. In the optimization problem, an adaptive balance parameter is designed to trade off the overall accuracy and redundancy. Additionally, we adopt an unsupervised overall accuracy prediction method to obtain the overall accuracy; thus, no ground truth or training samples is required. Experimental results on the ROSIS and RetigaEx data sets show that our method outperforms four representative methods in terms of classification accuracy and redundancy.