Sparsity Score Estimation for Hyperspectral Anomaly Detection

Hyperspectral image usually possesses complicated conditions of land-cover distribution, which brings challenge to achieve an effective background representation for hyperspectral anomaly detection. Sparse learning gives a way to overcome this obstacle. A novel sparsity score estimation framework for hyperspectral anomaly detection (SSEAD) is proposed in this paper. Firstly, an overcomplete dictionary and corresponding sparse code matrix are obtained from the hyperspectral data. Then, the frequency of each dictionary atom for reconstruction, which is also called the atom usage probability (AUP), are estimated from the sparse code matrix, from which the strength of each atom for reconstruction is obtained. Finally, the estimated frequencies are transformed to the sparsity score for each pixel. In the proposed detection framework, two operations which aim to enhance the learned dictionary to be more effective for anomaly detection are implemented: 1) dictionary based background feature transformation, and 2) dictionary iterative reweighting. Two real-world hyperspectral datasets are utilized to evaluate the performance of the proposed framework. The experimental results show that the proposed framework achieves superior performance relative to some of the other state-of-the-art anomaly detection methods.