Hyperspectral band selection based on evolutionary optimization

A hyperspectral image consists of a series of spectral bands which has brought great challenges to image processing and analysis. To alleviate the curse of dimensionality, band selection is therefore applied to the hyperspectral images. In this paper, a two-step method is proposed for band selection. In the first step, the band selection is converted to a global optimization problem and tackled by evolutionary optimization. To this end, a new fitness function is designed as the optimization objective and a differential evolution (DE) algorithm is employed to optimize the objective and find the optimal bands. In the second step, a simplified optimum idea factor (SOIF) is used for a fine selection. The K-nearest neighbor(KNN) and support vector machine (SVM) classifiers are then used to evaluate the obtained bands. The experiment on the AVIRIS images demonstrates that our approach is more effective than some state-of-The-art methods.