Hyperspectral image classification based on spectra derivative features and locality preserving analysis

High spectral resolution and correlation hinders the application of traditional hyperspectral classification methods in the spectral domain. To address this problem, derivative information is studied in an effort to capture salient features of different land-cover classes. Two locality-preserving dimensionality-reduction methods-specifically, locality-preserving nonnegative matrix factorization and local Fisher discriminant analysis-are incorporated to preserve the local structure of neighboring samples. Since the statistical distribution of classes in hyperspectral imagery is often a complicated multimodal structure, classifiers based on a Gaussian mixture model are employed after feature extraction and dimension reduction. Finally, the classification results in the spectral as well as derivative domains are fused by a logarithmic-opinion-pool rule. Experimental results demonstrate that the proposed algorithms improve classification accuracy even in a small training-sample-size situation.