Some Techniques for Anomaly Detection in Hyperspectral Imageries

This research focuses on the very first step in the analysis of an image, the point at which one assumes no prior knowledge about the statistical characteristics of the pixels in the image and where little or nothing is known about the size and shape of the objects to be detected. Therefore, the only available option is to look for a point (or group of points) that deviates so much from other points as to arouse suspicion that it was generated by a different mechanism. This project does that by looking for one-dimensional projection (projection pursuit) optimizing some measurement of interest (index). Following work performed by the authors in visualization techniques for anomaly detection combining low components of PCA and RX (Alonso and Malpica, 2009), this work analyzes and compares index skewness and kurtosis with the popular RX index. The optimization for the plane projection is performed with a genetic algorithm. These indexes are tested in synthetic image and in AHS hyperspectral imagery. The current project shows how these indexes have their properties and characteristics and how they are superior to RX in many ways. Further, the authors proposed visualization techniques for anomaly detection combining the different indexes.