A Spatial–Spectral Combination Method for Hyperspectral Band Selection

Hyperspectral images are characterized by hundreds of spectral bands and rich information. However, there exists a large amount information redundancy among adjacent bands. In this study, spatial–spectral combination method for hyperspectral band selection (SSCBS) is proposed to reduce redundancy. First, the image automatically divided into subspaces. Seven algorithms classified as four types executed compared. The means algorithm most suitable subspace division input image, with calculation being fastest. Then, each subspace, adopted select best band. maximum more prominent characteristics between selected. parameters Euclidean distance angle used measure intraclass correlation interclass specificity, respectively. Weight coefficient quantifying intrinsic relationship pixels constructed, then optimal selected weight coefficients entropy. Moreover, an automatic in paper provide appropriate number sets, which out consideration existing research. experimental results show, compared other competing methods, that SSCBS approach has highest classification accuracy on three benchmark datasets takes less computation time. These demonstrate achieves satisfactory performance against state-of-the-art algorithms.