Statistical analysis approaches often are applied to output parameters of physics-based, target-fitting algorithms to improve the ability to distinguish intact unexploded ordnance (UXO) from metallic scrap. Although 100% detection of the UXO targets on fairly uncomplicated ranges can be approached, the current classification capability and resulting false alarm rates mean that clearing the ranges requires excavating 5 to 25 non-UXO targets to recover each intact UXO. Successful detection and classification of buried targets depends heavily on the quality of data available from the geophysical survey. Data density, motion noise, and navigation errors are a few of the many factors that govern data quality, but the relationship between these factors and operational survey practices (e.g., lane spacing, sensor configuration, and vehicle speed) are complex and not well understood. This project will apply the natural advantages of Monte Carlo simulation to allow precise investigation of these relationships without loss of flexibility in defining the behavior of each system component. This project aims to demonstrate proof-of-principle that a carefully constructed Monte Carlo simulator, combined with accurate models, can provide useful information to personnel directing UXO site cleanup activities for improving results of existing geophysical survey procedures.

The proof-of-principle demonstration will be performed for a single electromagnetic induction (EMI) sensor on one platform. The tool will include an accurate forward model to generate synthetic EMI signals, a realistic noise model, and a faithful application of existing data inversion and classification schemes. In a straightforward implementation of the classic Monte Carlo approach, researchers will generate statistics that reflect overall system performance for a few general baseline cases that represent existing survey procedures, as well as several other cases that represent different operational choices. Results from the recent SERDP project MR-1313 investigating the inherent variability of UXO targets will be incorporated into this research project. This project's results will be expressed chiefly in terms of probability of target detection (Pd) and probability of target classification (Pc).

Research will be closely coordinated with UXO cleanup directors in the U.S. Army Engineering and Support Center, Huntsville and will provide them with guidance in operational procedures. In addition, the methodology and models developed in this project can be used in designing EMI coils, developing operational procedures to address overlapping targets, and developing improved classification procedures.