Remote sensing image feature selection based on α-torrent rough set theory

Supervised classification in remote sensing imagery is receiving increasing attention in current research. In order to improve the classification ability, a lot of spatial-features have been utilized. Unfortunately, too many features often cause classifier over-fit to a certain features´ character and lead to lower classification accuracy. Feature selection algorithms have utilized to select useful feature and improve classification accuracy. Rough set theory, as a powerful analysis tool, has been proven to be effective in remote sensing classification field. But spectral uncertainty or vagueness caused by spectral confusion between-class and spectral variation within-class leads to the overlap in a large number of features. In these cases, the traditional rough sets can not perform effectively. To solve this problem, this research proposed a new feature selection method based on α-Torrent rough set theory. The experiments showed, compared with PCA and traditional rough set method, that our method could select usefully features and improved classification accuracy.