Subsurface electromagnetic induction imaging for unexploded ordnance detection

Detection and classification of unexploded ordnance based on electromagnetic induction have made tremendous progress over the last few years, to the point that not only more realistic terrains are being considered but also more realistic questions – such as when to stop digging – are being posed. Answering such questions would be easier if it were somehow possible to see under the surface. In this work we propose a method that, within the limitations on resolution imposed in the available range of frequencies, generates subsurface images from which the positions, relative strengths, and number of targets can be read off at a glance. The method seeds the subsurface with multiple dipoles at known locations that contribute collectively but independently to the measured magnetic field. The polarizabilities of the dipoles are simultaneously updated in a process that seeks to minimize the mismatch between computed and measured fields over a grid. In order to force the polarizabilities to be positive we use their square roots as optimization variables, which makes the problem nonlinear. The iterative update process guided by a Jacobian matrix discards or selects dipoles based on their influence on the measured field. Preliminary investigations indicate a fast convergence rate and the ability of the algorithm to locate multiple targets based on data from various state-of-the-art electromagnetic induction sensors.

► We propose a subsurface imaging algorithm in the electromagnetic induction regime.
► We implement the method and test it on both synthetic data and sensor data.
► We show examples of detection of single and multiple unexploded ordnance.
► We apply the method to new-generation electromagnetic induction sensors.