Neural Network and Support Vector Machine Classification of UXO Using Magnetics Finite Element Modeling Data

Abstract

A recently completed Montana Tech study used finite element modeling to calculate magnetic dipole moments for realistic unexploded ordnance (UXO) shapes found at Montana Army National Guard training sites. The modeling approach compared predictions from a simple prolate spheroid model, typical for conventional modeling, with a complex shape constructed using computer aided design (CAD). The complex shapes embody all of the features of the actual UXO bodies. Results showed that the larger, more accurate dipole moments calculated from the complex shape can be used to adjust the dipole moment discrimination level, ultimately reducing the number of targets to be dug.