Automatic Detection of UXO from Airborne Magnetic Data Using a Neural Network

Abstract

Recent developments in airborne magnetic detection systems have made it possible to detect small ferro-metallic objects. However, airborne magnetic data can be really large and, therefore, there is an increasing need for a fully automatic interpretation technique that could be used to make decisions regarding the nature of the sources in the field in real time. The massively parallel processing advantage of artificial neural networks makes them suitable for hardware implementations; therefore, using these networks in conjunction with a magnetic system has the potential to greatly speed up the detection of ferro-metallic objects. In this paper, we have developed a new method for detection and characterization of unexploded ordnance (UXO) using a Hopfield neural network as applied to airborne magnetic data. The Hopfield network is used to optimize the magnetic moment of a dipole source representing the magnetic object at regular locations. For each location, the Hopfield network reaches its stable energy state. The location of the object corresponds to the location yielding the minimum Hopfield energy. Output results include position in two dimensions (horizontal location and depth), magnetic moment, and effective inclination. Theoretical and actual field examples show that the Hopfield network is accurate and objective tool for the detection of UXO. Moreover, because the Hopfield neural network is a natural analog-to-digital converter, it is ideally suited for incorporation into airborne magnetic instrumentation systems.