Spatial -Spectral Feature Extraction using Three-Dimensional Singular Spectrum Analysis for Hyperspectral Image Classification

Feature extraction has a valuable role in hyperspectral images processing. In recent years, various methods have been presented to extract efficient features of hyperspectral images. Recent studies have successfully used conventional singular spectrum analysis in the spectral domain and two-dimensional singular spectrum analysis in the spatial domain for feature extraction in hyperspectral images. However, a lack of success in joint spectral-spatial feature extraction is a problem with both algorithms. This study uses a three-dimensional singular spectrum analysis extension to overcome this problem. The implementation of proposal model on hyperspectral images removes the noise components during spectral-spatial feature extraction process and significantly improves features identification capability. This study conducts experiments using two publically available datasets. Experimental results show that our proposed method has a promising performance so that it has obtained a classification accuracy of at least 1.93% and 1.27% respectively on the hyperspectral dataset of Indian Pines and Pavia University compared to other recent methods.