A Fast Hyperspectral Classification Method by Integrating Rotational Invariant Linear Discriminant Analysis and Nearest Regularized Subspace

Abstract

Although feature extraction methods can improve hyperspectral image (HSI) classification speed, how to improve the effectiveness of HSI classification is also still a big challenge due to its high spectral dimensionality. The classical feature extraction methods increase the speed of classification at the cost of accuracy lost. By analyzing the correlation and differences between classes in the training samples, a new hyperspectral classification method called RILDA-NRS based on rotational invariant linear discriminant analysis theory (RILDA) and nearest regularized subspace (NRS) theory was proposed in this paper. RILDA was first used in the proposed method to extract the useful spectral features from hyperspectral images, in which not only the dimensionality of HSI is reduced, but also the discriminability between samples is enhanced. Then, the feature extraction results are embedded into the NRS classification model to classify HSI. The experimental results have demonstrated that the proposed method has obvious advantages in terms of classification accuracy and speed.