Band selection for hyperspectral images based on particle swarm optimization and differential evolution algorithms with hybrid encoding

Abstract

Evolutionary algorithms have been widely used in band selection for hyperspectral images. The particle swarm optimization (PSO) and the differential evolution (DE) algorithms are two common evolutionary techniques with efficient optimization capabilities. In order to fully utilize the advantages they provide, a band selection method is proposed based on the two algorithms with hybrid encoding. This method firstly uses hybrid encoding to make PSO and DE suitable for band selection. Secondly, the classification accuracy of an SVM classifier is used as the fitness function. Thirdly, we adopt the double population parallel iterative method to search for the optimal band combination. The experimental results on AVIRIS hyperspectral data show that the average classification accuracy of our proposed method is higher than the binary PSO algorithm, higher than the hybrid particle swarm algorithm, and higher than the hybrid coding differential evolution algorithm. These classification results demonstrate the effectiveness of the proposed method.