Band Selection For Hyperspectral Data Based On Clustering And Genetic Algorithm

Band selection is the most useful method to overcome data redundancy of hyperspectral image data, which is to find a subset of spectral bands from more than 100 original spectral bands. In this chapter, we make a short review of band selection methods developed in recent years, and a new band selection method based on k-means++ clustering and genetic algorithms is developed. Firstly, k-means++ clustering method is applied to generate several optional subsets of spectral bands from original hyperspectral data. Then genetic algorithm is used to select bands based on a framework of supervised classification of the hyperspectral data. Two hyperspectral datasets are used to test the proposed method. It is proved that the presented band selection method can reduce the dimensionality of hyperspectral data efficiently and result in much better accuracy for classification applications.