Anomaly detection and estimation in hyperspectral imaging using Random Matrix Theory tools

Anomaly detection aims to detect sources with different spectral characteristics from the background in an hyperspectral image. Classical tools for anomaly detection and estimation are known to have poor performance when they are used on high dimensional hyperspectral image since typically both the number of available sample and their size are large for this kind of imaging. New estimation methods for the number of anomalies, adapted to large dimensional systems, are required. This article points out the limits of classical methods such as Akaike Information Criterion (AIC) or Minimum Description Length (MDL) criteria and it proposes a new estimator based on Random Matrix Theory results better adapted for hyperspectral imaging. Finally, the proposed method is validated on both Monte-Carlo simulations and on experimental data.