Self-weighted collaborative representation for hyperspectral anomaly detection

Anomaly detection has become an alluring topic in hyperspectral imagery (HSI) processing over the last ten years. Recently, the collaborative representation-based detector (CRD) has been proposed and shows good detection performance for hyperspectral imagery. However, the original CRD assumes that the importance of each band are equal, which is not pragmatic in practical application. To alleviate this problem, we propose a self-weighted collaborative representation-based detector (SWCRD) which combines the weight learning and collaborative representation into a joint objective function. The proposed SWCRD can assign suitable weights to each band and achieve collaborative representation simultaneously. Experimental results on two real hyperspectral datasets validate the outstanding detection performance of our proposed SWCRD compared with the original CRD.