Anomaly Detection for Hyperspectral Images Based on Robust Locally Linear Embedding

Abstract

In this paper, anomaly detection in hyperspectral images is investigated using robust locally linear embedding (RLLE) for dimensionality reduction in conjunction with the RX anomaly detector. The new RX-RLLE method is implemented for large images by subdividing the original image and applying the RX-RLLE operations to each subset. Moreover, from the kernel view of LLE, it is demonstrated that the RX-RLLE is equivalent to introducing a locally linear embedding (LLE) kernel into the kernel RX (KRX) algorithm. Experimental results indicate that the RX-RLLE has good anomaly detection performance and that RLLE has superior performance to LLE and principal component analysis (PCA) for dimensionality reduction in the application of anomaly detection.