In recent years, survey data analysis and processing techniques for use with commercial sensors have been developed that improve detection capabilities and discrimination between unexploded ordnance (UXO) and clutter. Among other things, these techniques include procedures for estimating target properties from survey data (e.g., size, shape, orientation, etc.) and feature-based classification procedures for discriminating between different types of targets. These feature-based processing procedures, when used with carefully collected survey data, have been shown during multiple technology demonstrations to significantly improve UXO and clutter discrimination over that achieved using traditional procedures.

This project integrated UX-Analyze, federally-funded processing and analysis tools, into Oasis Montaj™, commercially available survey-data visualization and processing software. In addition, the project validated the procedures using data acquired at field sites and documented the realized cost and performance improvements associated with feature-based UXO characterization.

Technology Description

During the course of several SERDP and ESTCP projects, AETC and Duke University (partners in this program) developed various advanced processing procedures for improved detection of buried UXO and discrimination between UXO and clutter. Different procedures have been developed for use with magnetometer data and electromagnetic induction (EMI) sensor data. The procedures rely on physics-based models for the sensor response due to buried objects and estimate model parameters that correlate with target features or location to produce an optimal match between the modeled- and measured-sensor response. Target location, depth, and magnetic dipole moment can be determined from magnetometer survey data, and the size of the target can be estimated from the dipole moment. These parameters have proven to be useful for discriminating between buried UXO and some clutter items. With proper processing, EMI sensor data collected above an unknown object can be used to determine eigenvalues of the magnetic polarizability tensor, which in turn can be used to determine information regarding the object's shape, size and burial depth. Robust, statistically efficient decision rules for target classification and discrimination then can be constructed using any or all of the target features. In addition to the anomaly characterization and classification methods discussed, the following software tools were included to support data analysis: (1) quantitative inversion of data acquired using arrays of EM61 sensors, (2) auto-leveling of multi-channel EMI data, (3) magnetic soil and metal discrimination algorithms, and (4) quantitative inversion of EM63 data using model-based parametric techniques.

Demonstration Results

Multiple technology demonstrations were conducted to support this project at sites including Aberdeen Proving Ground, Maryland; Lake Success Business Park, Connecticut; and Badlands Bombing Range, South Dakota. In addition, this project participated in a demonstration at the former Camp Sibert, Alabama which was the first in a series of ESTCP live site demonstrations to validate the application of a number of recently developed classification technologies in a comprehensive approach to munitions response. Based on this project’s demonstrations, the processing and analysis tools were successfully validated to support classification. Additionally, a Description and Features of UX-Analyze technical report was developed that summarizes the logic, features, and operations of UX-Analyze.

Implementation Issues

The most significant benefit of this effort is the ability to estimate target features and create a feature-based prioritized list for geophysical anomalies using commercially available processing software. Ancillary benefits result from the front-end pre-processor that will evaluate survey data quality with respect to the requirements for successful processing and the back-end statistical analysis that will determine which set or sets of features provide the optimal classification performance.