The objective of this research project was to develop an improved understanding of the fundamental physics underlying classification and discrimination of unexploded ordnance (UXO) and clutter items using electromagnetic induction (EMI) data. This improved understanding can help to establish a rational basis for the development and implementation of improved UXO/clutter discrimination processing and analysis techniques capable of fully exploiting the capabilities of new EMI sensor technologies purpose-built for target classification.
EMI signature data were collected for inert munitions and clutter items recovered from four of the ESTCP Classification Pilot Program Live Site Demonstrations sites (former Camp San Luis Obispo, California; former Camp Butner, North Carolina; former Camp Beale, California; and former Pole Mountain Target and Maneuver Area, Wyoming). Signature data for items from the former Camp Sibert, Alabama demonstration were collected under the previous SERDP project MR-1595. The project team also collected signature data for a variety of clutter items recovered following the TEMTADS man-portable adjunct (ESTCP project MR-200909) demonstration at the former Remington Arms site in Bridgeport, Connecticut, along with miscellaneous inert munitions from various sources. All of the data were inverted using the standard dipole response model to obtain principal axis polarizabilities.
The EMI response measurements were made in air with the targets supported on foam blocks below the TEMTADS array. Background (no target) signal levels were periodically measured during data collection and subtracted from target-present measurements. Because of the extremely large conductivity contrast between soil and typical munitions and clutter (approximately 1 S/m vs. 106 S/m), target signatures measured in air are indistinguishable from the signatures of objects buried in the ground.
The data were processed to obtain principal axis polarizabilities using standard dipole inversion. The basic idea is to search out the target location, orientation, and principal axis polarizabilities that minimize the difference between the measured responses and those calculated using the dipole response model. With some variations in the details, this approach has been used by a number of research groups in developing procedures to classify buried objects on the basis of their EMI response.
For the Naval Research Laboratory array data, inversion was accomplished by a two-stage method. In the first stage, the target’s (X, Y, Z) dipole location beneath is solved for non-linearly. At each iteration within this inversion, the nine-element polarizability tensor is solved linearly. It is required that this tensor be symmetric; therefore, only six elements are unique. Initial guesses for X and Y are determined by a signal-weighted mean. The routine loops over a number of initial guesses in Z, keeping the result giving the best fit as measured by the chi-squared value. The non-linear inversion is conducted simultaneously over all time gates, such that the dipole (X, Y, Z) location applies to all decay times. At each time gate, the eigenvalues and angles are extracted from the polarizability tensor.
In the second stage, six parameters were used: the three spatial parameters (X, Y, Z) and three angles representing the yaw, pitch, and roll of the target (Euler angles y, q, f). Here the eigenvalues of the polarizability tensor are solved for linearly within the six-parameter non-linear inversion. In this second stage, both the target location and its orientation are required to remain constant over all time gates, consistent with the basic dipole response construct. The value of the best fit X, Y, and Z from the first stage, and the median value of the first-stage angles are used as an initial guess for this stage. Additional loops over depth and angles are included to better ensure finding the global minimum.
This project assembled a substantial library of EMI signatures of munitions and clutter which has proven to be useful for research purposes and to support development of classification procedures. As new, site-specific inert munitions and clutter items become available through the ESTCP live site demonstrations, they should be measured and included in the library.
The results indicate that a decay time interval of about 0.1 ms to 3 ms is adequate for classification. All of the data in the library were collected with decays to 25 ms. Some signals are distorted by run-on effects with shorter transmit pulses, so a degree of caution should be exercised when using the library with shorter-decay data.
Typical library matching procedures used today only ask how much a given target looks (in the EMI sense) like a target of interest. There is now an extensive library of polarizabilities not only for a variety of representative munitions items but also thousands of clutter items recovered from a number of different munitions response sites. Using this database, it is possible to evaluate an unknown target as more munitions-like or more clutter-like, which should improve classification performance. The library-based classification schemes using this approach should be studied in order to determine whether or not they offer improved performance.