DCT MFA subspace hyperspectral remote sensing images terrain classification

Hyperspectral remote sensing images (HRSI) terrain recognition encounters the difficulties of having large sample dimension and lacking samples with class labels, which cause the curse of dimensionality problem. Thus, feature extraction is demanded to reduce data dimensionality ahead of terrain recognition. A novel hyperspectral remote sensing images feature extraction algorithm, named as DCT MFA, is presented in this paper. Firstly, take discrete cosine transform (DCT) of each pixel spectrum vector and get the coefficients of DCT; and then, execute marginal Fisher analysis (MFA) for the coefficients of DCT and get the DCT MFA subspace. Minimum Euclidean distance classifier is used to evaluate the performance of extracted features. The recognition results on three real airborne-spaceborne visible-infrared imaging spectro-meter (ASVIRIS) HRSI prove that, compared with the spectrum linear discriminant analysis (LDA) subspace algorithm, the presented DCT MFA subspace approach does obviously increase the accurate identification rate.