Feature Extraction Using Multidimensional Spectral Regression Whitening for Hyperspectral Image Classification

Hyperspectral images (HSIs) consist of hundreds spectral bands, which can be used to precisely characterize different land cover types. However, an HSI has redundant information and is prone the “dimensionality curse.” Therefore, it necessary reduce through dimensionality reduction (DR), given that dimensions contain unique primary feature information, complementary. Accordingly, a new extraction method based on multidimensional regression whitening (M-SRW) proposed, reduces reconstructs for extraction. The proposed consists following steps: First, original superpixel segmented by entropy rate segmentation algorithm. Second, SRW performed in each block dimension dimension. Third, blocks same are combined obtain reconstructed HSI. Finally, support vector machine utilized classify dimensions, majority voting decision fusion final classification result map. Experiments three public hyperspectral data sets demonstrated M-SRW superior several state-of-the-art approaches terms accuracy.