The cost of identifying and disposing of unexploded ordnance (UXO) on closed and closing ranges in the United States using current technologies ranges from $8 billion to $35 billion. Typical survey methods use algorithms aimed at discriminating UXO from clutter in order to avoid the costs of excavating nonhazardous material. These discrimination algorithms all involve some form of comparison between observed signals produced by sensors and predicted signals produced by models. When a clear match is found, target identification occurs. In general, success is limited by noise in the sensors and error in the models. For electromagnetic induction (EMI) sensors, the interaction of target position with sensor position is typically described using the Dipole Model, whereby target response is the field of a point dipole located at the center of the target. This is a good approximation in cases where target dimensions are small relative to burial depths, but it begins to break down as burial depths become shallower, causing degradation of discrimination performance.

The objective of this project is to develop an empirical method for removing errors in the standard Dipole Model. Researchers will demonstrate this method for one UXO type, an 81-millimeter (mm) mortar. Under SERDP project MR-1121, the problem of non-dipole model effects was identified and partially addressed using a semi-analytical approach. This project will develop a straightforward, purely empirical approach to improve on those results.

Technical Approach

In this project, researchers will measure EMI response of a single 81-mm mortar at a range of depths and orientations, fit the Dipole Model to these data, then calculate the empirical correction factors needed to eliminate errors. The project team previously has observed that these correction factors vary smoothly with position and orientation, indicating they can be accurately represented with splines. The approach is to develop multidimensional splines and demonstrate that they efficiently remove error in the Dipole Model. Each UXO type would require two splines, one for longitudinal correction and one for transverse correction. The splines themselves will be constructed and queried using existing routines from numerical recipes.


Benefits of this project relate to the accuracy of fitting EMI survey data for the purpose of discriminating UXO from clutter. This project will demonstrate improved discrimination for the 81-mm mortar due to the correction for non-dipole effects. When expanded to additional UXO types, this project will remove the inherent modeling error that occurs when the Dipole Model is used to approximate the response of real UXO and will improve the basis for successful deployment of discrimination schemes based on EMI technology. Overall, this research is directed at improving the probability of detection while minimizing false alarms, thereby improving the receiver operating characteristics (ROC) curve and leading to cost savings.