Hyperspectral anomaly detection based on collaborative representation of dictionary subspace

Traditional methods usually show low detection performance, which is caused by the inaccurate evaluation of background statistical characteristics and the contamination of the anomaly. In response to these problems, this paper proposes a novel anomaly detection method based on collaborative representation of dictionary subspace. Since the prior information of the background and the anomaly is not known, the proposed method first utilizes mean shift method to cluster the original hyperspectral image (HSI). Then, some pixels in each category are selected as background dictionary atoms. Therefore, an representative overcomplete background dictionary can be obtained in this way. This dictionary can avoid the contamination of most anomaly pixels. Finally, collaborative representation is performed to detect anomaly targets by using the dictionary instead of the domain information of the pixel under test. It does not require modeling the background, which improves the accuracy of the calculation. Experimental results show that the proposed method outperform the other state-of-the-art methods.