Consensus Anomaly Detection Using Clustering Methods in Hyperspectral imagery

A common anomaly detection algorithm for hyperspectral imagery is the RX algorithm based on the Mahalanobis distance of each pixel from the expected value of that pixel. This algorithm can be applied either directly on a hyperspectral image or on a dimensionality-reduced hyperspectral image. Recent work on Non-Negative Matrix Factorization (NNMF) provides a fast-iterative algorithm for decomposing a hyperspectral cube and achieving dimensionality reduction. In this paper, we present the RICHARD (Robust Iterative Consensus Anomaly RX Detection) algorithm that generates more than 100 RX tests after data manipulations (such as Principal Component Analysis (PCA) and NNMF) which vary in their specific parameters; we then use a weighted consensus voting process in order to detect anomalies without any prior knowledge. Using the RICHARD algorithm can enhance our options in finding obscure anomalies which do not appear in every algorithm.