A morphological approach to automatic mine detection problems

We consider the problem of detecting mines and minelike targets imaged by multispectral sensors. We propose an algorithm, based on mathematical morphology (MM), that yields accurate detection results in moderately cluttered environments. For targets in heavily cluttered environments, a preprocessing step is employed, based on the maximum noise fraction (MNF) transform, in order to reduce the effect of clutter and enhance the presence of targets. The algorithm is simple, performs well, and requires only approximate knowledge of target size.