Detection of Land Mines in Multi-Spectral and Multi-Temporal IR Imagery

In this paper we present techniques for detecting mines using data acquired at multiple wavelengths (multi-spectral) or at multiple times (multi-temporal). Both types of data employ vector-valued pixels, and therefore they have a common mathematical structure. For the multi-spectral data we show that the performance of standard algorithms can be improved signiÞcantly through spatial whitening when there is spatial correlation in the scene. For multitemporal data we employ a classiÞer of the vector-valued pixels to identify mine-like and clutter-like regions. Both multi-spectral and multi-temporal algorithms are demonstrated using experimental data collected at Fort A.P. Hill, VA under a variety of conditions.