Detection of hyperspectral anomalies using density estimation and collaborative representation

摘要

The collaborative-representation-based detector (CRD) will misjudge the anomaly pixel under test as a background pixel if there are a few anomalies similar to the pixel under test in the original background. In order to solve the problem, a density-estimation-based background refinement method is proposed to remove the anomalies from the original background. In the method, anomaly degree for each pixel in the original background is estimated by calculating its probability density. A smaller probability density indicates that the pixel has a larger anomaly degree in a background area. Then, pixels with larger anomaly degree are removed from the original background via Otsu's method. Finally, the refined background is combined with collaborative representation method to detect the anomalies among the hyperspectral imagery. To validate the effectiveness of the proposed algorithm, experiments are conducted on real hyperspectral dataset. The results show that the proposed algorithm performs better compared with the current anomaly detection algorithms.