Hyperspectral Image Endmember Extraction Algorithm Based on Manifold Learning and Superpixel Feature Extraction

Endmember extraction is the crucial process of hyperspectral unmixing. In this study, we introduce a nonlinear endmember extraction algorithm for hyperspectral remote sensing image analysis. The algorithm use the manifold learning technique to acquire the intrinsic nonlinear structure of hyperspectral data, and extract the endmembers based on convex simplex theory in low-dimensional manifolds. In order to achieve better manifold representations and reduce the computational burden, superpixel feature extraction is performed on the original hyperspectral data. Experimental results on synthetic and real data demonstrate the effectiveness of the proposed method.