Band selection for hyperspecral imagery based on particle swarm optimization

Abstract

An effective band selection method can greatly improve both the speed and effect of hyperspectral images process. The Minimum Noise Band Selection (MNBS) method is able to find subset of bands with high SNR and low correlation. Combine with MNBS, a new band selection method called PSO_MNBS based on particle swarm optimization (PSO) is proposed. Firstly, the number of selected bands is estimated by a virtual dimension (VD) method. In order to obtain better and stable result, the result of MNBS implemented by sequential forward selection is used to construct initial particle swarm. The optimal band subset is then obtained by PSO based on MNBS criterion. The proposed method is evaluated by experiments of hyperspectral anomaly detection. The results demonstrate that PSO_MNBS can effectively select the most significant bands for hyperspectral anomaly detection and shorten the CPU time at the same time.