Neural Network Classification Techniques for UXO Applications

A neural network was used to predict the mass of shallow subsurface conductive objects. The field data was from an unexploded ordnance (UXO) survey conducted at Camp Simms, Washington DC. The survey instrument was the Geonics EM-6 1 pulsed induction sensor. The purpose of the study was to develop a neural network architecture that could be fielded at numerous sites under the US Army Corps of Engineer’s Ordnance and Explosives Knowledgebase (OE-KB) program. The neural network was successful in predicting masses and depths. This paper presents the results of a single training session and is also intended as a short tutorial in how to prepare and present geophysical field data for analysis by a neural network.