Dimensionality reduction of hyperspectral data based on centroid feature

Hyperspectral data consists of large number of images in narrow contiguous wavelength bands. In To reduce dimension of data, centroid amplitude coordinate of area under the spectral response curve (SRC) has been proposed, in this study, as feature. The methodology is based on dividing the area under SRC into subsets and calculating the centroid amplitude coordinate of each subset and thus, getting dimension reduced. Optimum features have been used for classification and accuracy assessment of (Anderson´s) higher level land cover (nine) classes from AVIRIS 220 band Indian Pine (USA) huperspectral data. An overall classification accuracy of 86.30% has been achieved by using features based on spectral response in the Green, Red, Very Near Infra Red and Short Wave Infra Red bands.