Self-paced collaborative representation with manifold weighting for hyperspectral anomaly detection

In the acquisition process of hyperspectral images (HSIs), each band may be contaminated with different degrees of mixing noise. For hyperspectral anomaly detection (HAD) tasks, bands with higher noise contamination levels provide more interference information, thus affecting the detection results. In order to reduce the negative effect of noise in HSIs and improve the robustness of the detector, we propose a self-paced collaborative representation with manifold weighting hyperspectral anomaly detector (SPCRMW). Each band is given an order to join the collaborative representation model according to the degree of noise contamination. Moreover, a novel manifold learning reconstruction-based regularization matrix is proposed to reduce the effect of potential anomalous pixels mixed in the background on collaborative representations. It can automatically assign weights to the background pixels by manifold learning reconstruction error. The results compared with six state-of-the-art HAD methods on three real hyperspectral datasets are presented and illustrate the superiority of the proposed SPCRMW method.