No abstract

This paper investigates the ability of a supervised artificial neural network to estimate the location of buried steel drums from magnetic anomaly data. The neural network was trained using magnetic signatures of a magnetic dipole source, which experience shows to be equivalent to the field from a buried drum. It is demonstrated that for sources buried at depths less than [Formula: see text], the supervised neural network estimates the source locations from theoretical magnetic data with errors not exceeding [Formula: see text] for depth and [Formula: see text] for horizontal location. The network also showed a potential to locate the sources in the presence of noise. The technique was also tested on field data collected at a Stanford University test site. The estimated horizontal locations and depths from field data collected over buried steel drums are close to the ground truth, with average errors of [Formula: see text] for depths and [Formula: see text] for horizontal locations. These estimated depths are better than those given by the commonly used Euler method, whereas the horizontal locations are comparable. Using the trained network, the solutions could be derived in a few milliseconds. The time required for training is significant, but could be reduced with faster computers.

No abstract

No abstract