Band selection based on evolution algorithm and sequential search for hyperspectral classification

Band (feature) selection for multispectral or hyperspectral data is an effective method to reduce dimension for cutting down the computational cost and alleviating the Hughes phenomenon. An efficient feature selection method based on evolution algorithm (PSO and GA) and sequential search is proposed. The method embeds the sequential search into the evolution optimization for better ability of the fine tune in local search space and thus behaves well in both global and local cases. In addition, the embed scheme guarantees the validity of solutions for the 1-st form feature selection problem. The experiments with an airborne visible/infrared imaging spectrometer (AVIRIS) data set show the effectiveness of the proposed method.