25-10-2012, 04:48 PM
Innovative Statistical Inference for Anomaly Detection in Hyperspectral Imagery
ABSTRACT
A statistical motivated idea is proposed and its application to hyperspectral imagery is presented, as a viable alternative
to testing a two-sample hypothesis using conventional methods. This idea led to the design of two novel algorithms for object
detection. The first algorithm, referred to as semiparametric (SemiP), is based on some of the advances made on
semiparametric inference. A logistic model, based on case-control data, and its maximum likelihood method are presented,
along with the analysis of its asymptotic behavior. The second algorithm, referred to as an approximation to semiparametric
(AsemiP), is based on fundamental theorems from large sample theory and is designed to approximate the performance
properties of the SemiP algorithm. Both algorithms have a remarkable ability to accentuate local anomalies in a scene. The
AsemiP algorithm is particularly more appealing, as it replaces complicated SemiP’s equations with simpler ones describing
the same phenomenon. Experimental results using real hyperspectral data are presented to illustrate the effectiveness of both
algorithms.