Feature Extraction Using Multidimensional Spectral Regression Whitening for Hyperspectral Image Classification

Abstract

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

Keywords

• Hyperspectral image (HSI) classification
• manifold learning
• multidimensional spectral regression whitening (M-SRW)
• unsupervised dimensionality reduction (DR)