Hyperspectral Images Terrain Classification in Combination Spectrum DLDA Subspace

Hyperspectral images face the problem of high dimensionality and low samples number, which results in unsatisfied recognition efficiency, thus dimensionality reduction is needed before terrain classification. A novel hyperspectral images feature extraction method is presented for dimensionality reduction. Firstly, take discrete Fourier transformation (DFT) of each pixel spectral curve, and combine the amplitude spectrum and corresponding phase spectrum; then direct linear discriminant analysis (DLDA) is performed in the combination spectrum space to extract features. Minimum distance classifier is used to evaluate the feature extraction performance in the achieved combination spectrum DLDA subspace. The experimental results for airborne visible/infrared imaging spectrometer (AVIRIS) hyperspectral image show that, comparing with the spectral DLDA subspace method, the present method can improve the terrain classification efficiency.