An improved adaptive band selecting method for hyperspectral image classification and dimension reduction

摘要

Band selection is an important problem in hyperspectral image processing. In this paper, an improved adaptive band selection method is proposed to choose the best bands combination which can provide useful information, small correlation, and good property for classification. The computing process is firstly conducted by block computation to select the more valuable bands. Secondly, the Euclidean distances between each band are considered to make the selected bands more different. At last, the classifying property of different objects is combined into the method so as to facilitate the classifying operation. Our experiment results show that the bands selected by this method could provide more valuable data for the oil spill image analysis.