Classification algorithms for landmine detection using hyperspectral imaging

In this paper we show some preliminary results obtained when applying various supervised classification algorithms for the detection of landmines in hyperspectral images. These classification algorithms already exist in the literature and are usually used for the general purpose of target detection using hyperspectral imaging. Landmines constitute a special case of targets because they are usually rare in the scene, may be covered by other materials (vegetation, sand, etc.) and they are made of different components so they have different reflectance spectra. In this study, we present the most suitable supervised classification techniques for the detection of landmines, their theoretical background and propose possible ways to improve their performance. The comparison between the classification techniques is based on three criteria: Probability of detection, False Alarm Rate and computation time.