The Standardized UXO Technology Demonstration Sites were established by the Environmental Security Technology Certification Program (ESTCP) under project MR-200103 to ensure that critical unexploded ordnance (UXO) technology performance parameters such as detection capability, false alarm rates, discrimination, reacquisition, and system efficiency are determined through standardized test methodologies, procedures, and facilities. This project performed standardized analyses of data collected at the two standardized sites at Aberdeen Proving Ground (APG) in Maryland and Yuma Proving Ground (YPG) in Arizona in order to understand the factors that affect UXO detection and discrimination. This project focused on electromagnetic induction (EMI) sensor and total field magnetometer systems.


The objective of this project was to implement standardized, data-level evaluations of demonstration performance at the Standardized UXO Technology Demonstration Sites These evaluations will support an understanding of the capabilities and limitations of the various UXO detection and discrimination sensors, as well as the influence of different soil conditions and field scenarios on the performance of those sensors.

Demonstration Results

Several survey data sets collected by demonstrators at the Standardized UXO Technology Demonstration Sites were analyzed using the standard dipole response model.  The magnetic or electromagnetic response of a target is represented by an induced dipole moment at the target location. Spatially mapped data collected over the target were inverted using this model to determine the target's location and depth and the parameters or features that characterize the target response. Performance metrics used in the analyses include target signal-to-noise ratio (SNR), dipole fit quality (squared correlation between the data and the dipole model fit to the data), and the accuracy of target depth and target parameter estimates (size, polarizabilities) used for target classification and discrimination. With carefully collected data, dipole fit quality increases with target SNR in a predictable way, so comparisons of dipole fit quality and SNR help to establish if poor fits are due to weak signals and/or high noise levels or to data collection problems. Target characterization and discrimination is based on the target parameters or features determined from inversion of the data using the dipole response model. Dipole fit quality is generally the best indicator of the fidelity of the calculated target parameters.

Target depth estimates from EMI data generally improved with fit quality. Most of the estimated depths are within ±25 cm, but even with the targets restricted to fit quality greater than 0.98 there are significant outliers. There are also significant outliers in the corresponding target size estimates, generally corresponding to targets for which the depth was overestimated. However, for the bulk of the data the target size estimates vary with UXO caliber as predicted by theory. Some classification and discrimination on the basis of target size may be possible at limited use sites where the UXO have a narrow size range. However, given the overall data quality in the surveys analyzed, shape-based discrimination is problematic. The researchers believe that the basic problem is geo-location errors, which add a substantial noise component to the dipole inversion andsignificantly degrade the accuracy of the estimated target features.

Results from the magnetometer analysis show that very accurate depth estimates are obtained with data having fit quality greater than 0.98. For these data, the size estimates are consistent with historical data. Analysis of remnant magnetization based on a comparison of results from degaussed and non-degaussed UXO targets shows no significant differences between target parameter distributions for the two classes.

Implementation Issues

Generally speaking the results are disappointing. Only a relatively small fraction of the survey data analyzed here is accurate enough to support reliable feature-based target classification and discrimination. Even when the target SNR is relatively high, the dipole fit quality is frequently relatively poor, suggesting that the problems are due to deficiencies in the surveys, especially sensor location errors.