Abstract
At present, metal detectors (MDs) are still widely used in mine detection tasks around the world. However, discrimination between metal fragments and landmines remains a difficult problem. This paper introduces a new processing method for MD signal dataset captured in the 3D surrounding space of a given target. A detailed investigation of such a signal dataset taken by a robotic arm, from various metal pieces with different shapes and sizes was firstly carried out for different postures and distance from the MD, and it was verified that the depth, metal shape, type and surface area can be successfully estimated. This basic methodology was then applied to examine a dataset taken from many real mines, and the results show that the probability of discrimination between metal fragments and types of landmines can be greatly improved.
