Anomaly Detection Based On Pca And Local Rxosp In Hyperspectral Image

Aiming at the noise vulnerability and the low detection performance of the classical RX algorithm under the complex background, an improved RX-OSP hyperspectral anomaly detection method is proposed. Firstly, PCA dimension reduction method is applied to suppress the background of hyperspectral image. Secondly, RX operator is used to detect the pixels owning the most prominent anomaly and the pixels are projected to their orthogonal complement subspaces. Then RXOSP processing is repeated according to the foregoing steps until there is no obvious anomaly. During the process of detection, the covariance matrix is calculated by localization instead of the traditional global approach to reduce the false detection effectively. Finally, ROC curve is adopted as the evaluation index for the experiment results, which reveals that the improved RXOSP algorithm is superior to RX, PCA-RX and RXOSP algorithms.