Truncated Nuclear Norm Based Hyperspectral Unmixing Method

In this study, we propose a semi-supervised hyperspectral unmixing method to obtain the abundances of the spectral signatures in a hyperspectral image. In the proposed model, low-rank representation (LRR) and a sparsity inducing norm is simultaneously hold within a sliding window of fixed size to capture the spatial structure within the adjacent pixels. Truncated nuclear norm (TNN), which is defined as the nuclear norm subtracted by the sum of the largest few singular values, is used to estimate the rank of abundance matrix within the window. We conduct several experiments on synthetic and real hyperspectral data sets to validate the effectiveness of the proposed method in obtaining the abundance maps.