Hyperspectral band selection based on a variable precision neighborhood rough set.

Band selection is a well-known approach for reducing dimensionality in hyperspectral images. We propose a band-selection method based on the variable precision neighborhood rough set theory to select informative bands from hyperspectral images. A decision-making information system was established by hyperspectral data derived from soybean samples between 400 and 1000 nm wavelengths. The dependency was used to evaluate band significance. The optimal band subset was selected by a forward greedy search algorithm. After adjusting appropriate threshold values, stable optimized results were obtained. To assess the effectiveness of the proposed band-selection technique, two classification models were constructed. The experimental results showed that admitting inclusion errors could improve classification performance, including band selection and generalization ability.