Feature selection method based on modified genetic algorithm with hyperspectral images

A new method based on genetic algorithm (GA) is proposed for feature selection of hyperspectral images. The proposed method fully uses the merit of genetic algorithm in parallel search and global optimization in terms of the application of feature selection of hyperspectral images. It exploits criteria that represent class separability to implement the individual evolution through crossover and mutation. To accelerate convergence and improve its performance, we introduce competition between two generations to simple genetic algorithm, and obtain the optimal combination of features for classification. The numerical experiments are performed on hyperspectral data with 200 bands collected by airborne visible/infrared imaging spectrometer (AVIRIS). The experimental results show that the proposed method has high classification accuracy and low computation cost for feature selection.