Weighted Anomaly Detection Algorithm Based on Local Density

Anomaly detection for hyperspectral imagery (HSI) has drawn a lot of attention lately. In the past decades, several methods for detecting anomalies in HSI have been developed. The Reed-Xiaoli detector (RXD) is one of the most widely used methods for anomaly detection. However, RXD assumes that background follow the Gaussian distribution, not always conformed to the real case. In order to solve this problem, we improve the RX anomaly detection algorithm by weighting changes of background. First, the concentric double-window model is utilized to obtain the local densities of each background pixels. Then, the local densities are applied to weight background mean and covariance matrix to reduce the influence of abnormal pixels in the background. The experimental results show that our method outperforms other approaches in anomaly detection.