Work to enhance the unexploded ordnance (UXO) classification ability of the Multisensor Towed Array Detection System (MTADS) has been based on making use of both the location information inherent in an item’s magnetometry response and the shape and size information inherent in the response to the time-domain electromagnetic induction (EMI) sensors in either a cooperative or joint inversion. In all of these efforts, classification performance has been limited by the information available from the EMI sensor. The industry-standard EMI sensor is the Geonics EM61, which is a time-domain instrument with either a single time gate to sample the amplitude of the decaying signal or four time gates sampling relatively early in time.

The objective of this project was to integrate an array of GEM-3 sensors with the MTADS platform and demonstrate the system’s classification performance.

To further improve UXO classification, a sensor providing more information was required. The Geophex GEM-3 sensor is a frequency-domain sensor with up to ten frequencies available for simultaneous measurement of the in-phase and quadrature response of the target. In principle, there is more information available from this sensor for use in UXO classification decisions. The commercial GEM-3 sensor is a handheld instrument not very amenable to rapid, wide area surveys.

A three-sensor array was designed and demonstrated at the Standardized UXO Demonstration Sites at Aberdeen Proving Ground (APG) and Yuma Proving Ground (YPG). Overall, the demonstration results were a disappointment. The GEM-3 array was capable of detecting ordnance to reasonable burial depths. All ordnance were detected on the APG Blind Grid to burial depths equal to 11 times the UXO item’s diameter. However, the system was not effective at distinguishing UXO from clutter. Even after adjusting for items emplaced too deep to be detected, high discrimination probabilities of detection (Pd) were achieved only at high levels of probabilities of false alarm (Pfa).

The primary cause is an issue of signal-to-noise ratio (SNR). Model-based, dipole inversions require fairly high levels of SNR (>10) to return reliable results.  The average noise levels varied from less than 1 ppm for mid-frequency quadrature responses to almost 10 ppm for the in-phase response at all frequencies. The signal level for most of the emplaced ordnance was in the range of one to tens of ppm. This response level is too weak to invert reliable ordnance signatures across all the frequencies for both the in-phase and quadrature responses. Distinguishing ordnance from clutter relies on accurate inversion of the in-phase/quadrature signature and the comparison of this signature to the library of expected ordnance. At low SNR, this is not possible.

Further complicating this library comparison process is the large number and variety of UXO items emplaced at the test sites. Given the error bars on the inverted parameters, there is necessarily a large overlap between UXO and clutter signatures.

It has also been noted that good inversion results from dynamic survey data require accurate positioning of the sensors, on the order of sub-cm positioning. The positioning and orientation of the GEM-3 array was provided by a cm-level Global Positioning System (GPS).

This project demonstrated that this system is not effective at distinguishing UXO from clutter. High discrimination Pd were achieved only at high levels of Pfa. Stakeholder acceptance of the use of discrimination techniques on real sites will require demonstration that these techniques can be deployed efficiently and with high Pd.