Band Selection Method of Hyperspectral Image for Classification Based on Particle Swarm Optimization

Hyperspectral remote sensing technology is a rapidly developing new integrated technology that is widely used in numerous areas. Rich spectral information from hyperspectral images can aid in the classification and recognition of the ground objects. However, there are some problems such as huge amount of data and high correlation between bands. Hence, the high dimensions of hyperspectral data must be reduced. The existing band selection algorithms have been analyzed in this paper regarding to the problems mentioned above. An application of particle swarm optimization (PSO) based on image information content and between-class separability criteria was proposed to band selection of hyperspectral image. The correlation coefficient matrix was formed for subspace partition. The fitness functions of PSO were reconstructed by using joint entropy as criterion of information content and Bhattacharya distance as between-class separability. Finally, the improved algorithm was tested with AVIRIS image data. The support vector machine (SVM) classification method was implemented to classify the selected optimal band combinations. Experimental results show that the proposed method can achieve higher classification accuracy than traditional methods.