Summary
In the past differentiating between unexploded ordnance (UXO) and clutter using shape information derived from conventional electromagnetic induction (EMI) data has proven challenging. Digital geophysical mapping using devices such as the Geonics EM61 showed some promise under certain scenarios but due to data deficiencies meaningful model parameters could not be reliably extracted.

To address this, research into advanced electromagnetic sensors specifically designed for UXO detection and classification began through funding by the US Department of Defense (DoD). The results of the research culminated in two commercially available timedomain electromagnetic sensors, the MetalMapper and the TEM2x2.

These new sensors allow the measured data to be inverted using a dipole model to produce consistent polarizabilities and decay parameters, as well as accurate location and depth of the buried object. These magnetic polarizabilities are intrinsic features that are used to classify a source as either UXO or clutter. The absolute and relative strength of the time-dependent polarizabilities relate to the object size and shape, whereas the rate at which they decay relates to the material properties and wall thickness. UXO are typically cylindrical, therefore they have one strong polarizability, and two weaker but equal polarizabilities. With these new advanced EM sensors and processes, reliable classification of suspected UXO targets using geophysical survey data is now possible, enabling millions of dollars in savings on cleanup projects.

Introduction
During an unexploded ordnance (UXO) or munition cleanup project, most of the items excavated are not UXO, but are harmless scrap metal. On several projects in the United States, it has been found that over 95% of the objects investigated were “clutter” – harmless metallic debris or geology (Andrews & Nelson, 2011). If items can be determined to be non-hazardous without digging, they can be left unexcavated, or can be excavated in a less costly way (i.e. without requiring a large exclusion zone or evacuations). Classification is the process of analysing data (in this paper, electromagnetic data) to decide whether each target is a hazard or not, and in some cases, even deciding which specific type of munition may be present.

Such classification (or discrimination) decisions, centred on physics-based analysis that is transparent and reproducible, will enable significant savings to be realized. This in turn will allow limited clean-up funds to effectively clear a greater number of areas.

Background
Classification as part of UXO surveys is nothing new; common field methods involve implicit discrimination. For example, using the common ‘Mag and Flag’ process (using hand-held detectors and placing pin flags where a strong signal is heard), decisions are made in the field based on the operator’s judgment.

In the 1980’s and 1990’s, the advent of digital geophysical mapping (DGM) methods provides the ability to archive and document the survey and the process of finding and selecting targets. DGM surveys are carried out using “second generation” sensors including high resolution magnetometers and electromagnetic (EM) devices such as the Geonics EM61. Interpretation of the results still involves professional judgement, but the results and decisions can be reviewed at any time. The target selection criteria are typically based on a combination of geophysical anomaly amplitude, footprint, and shape.

This is another form of classification, and may work to a limited extent if data collection procedures are appropriate. This data-feature centric interpretation is based on features, however, that are not necessarily related to features of the object. Furthermore, on sites where munitions and clutter items were of similar size to each other, no worthwhile classification was achieved by any of the analysts using EM61 data (Andrews & Nelson, 2011).

To address this, two US Department of Defense (DoD) agencies, the Strategic Environmental Research and Development Program (SERDP) and the Environmental Security Technology Certification Program (ESTCP) began funding research into “Third generation” sensors, specifically developed to facilitate the classification of UXO.

From this research, two are now commercially available through Geometrics Inc., the MetalMapper and the TEM2x2, both of which are time-domain electromagnetic (TEM) systems. These new sensors have three key differences from the previous (second generation) EM sensors:

  • Multiple transmitter and receiver coils in various
  • Orientations to measure multiple components at a
  • Single point in space,
  • Finer sampling of the time decay curve, and
  • Sample longer periods in time

These “third generation” sensors, result in more extensive
datasets, with ‘full illumination’ of a target.

Method
Classification of a target depends on calculating dipole model parameters that relate to intrinsic physical features of the objects rather than external features such as location and orientation of the source object. After checking data quality and preliminary processing, these model parameters are determined through forward modelling and inversion of the measured survey data, where the model parameters are continuously adjusted until a calculated solution is found that accurately reproduces the measured data.

This inversion process also yields an accurate location and depth of the buried object. The intrinsic features that are determined in this manner are referred to as magnetic polarizabilities. The absolute and relative strength of the time-dependent polarizabilities relate to the object size and shape, whereas the rate at which they decay relates to the material properties and wall thickness. UXOs are typically long cylindrical objects; therefore they have one strong (or primary) polarizability and two weaker (secondary/tertiary) but equal polarizabilities.

This is a very clear distinguishing characteristic of cylindrical objects.
Library matching techniques are then used to determine the similarity between the reference library measurements and the unknown sources (Figure 1). This approach has been found extremely successful in classifying UXO vs. nonhazardous objects, and in some instances being able to distinguish the type of UXO. Care must be taken to avoid misidentifying objects that are munitions not yet included in the library.

To meet this challenge, tools and an extensive quality assurance program has been developed, for example the placement of items in the survey area that are “blind” to the survey and interpretation teams. The objective of the classification process is to rank order the blind sources from most similar to the UXO signatures included in the library to least similar. While the primary objective is to identify UXOs, depending on the site, there can be requirements to also identify other items, for example rocket bodies and motors.

Figure 1. Example library match showing excellent similarity between the unknown source object (blue lines) and polarizabilities from a 37mm projectile (red lines).

Results
Since 2011, these new advanced EM sensors and processes have been evaluated at a number of live-site demonstrations that span a range of munitions types, vegetation, and terrain (SERDP-ESTCP, 2016). Each demonstration was conducted by one or more survey contractors on former or current US military installations where UXO are known to be present.

Once the survey has been completed, all anomalies are excavated to confirm each technology’s performance. Demonstrators were then scored based on their ability to eliminate non-hazardous items and identify all munitions. Results of the live site demonstrations clearly show that reliable classification of UXO using advanced EMI data can be routinely realized. Government program managers and regulators have been included in the demonstrations, and they are now beginning to require the use of UXO classification technologies in investigations on their sites.

Classification results from an example classification project using UX-Analyze software are shown in Figure 2 using a receiver-operator characteristic (ROC) curve. The ROC curve plots the results of all analyzed anomalies. The ROC curve is generated by stepping through the list in rank order and proceeding vertically if a UXO is encountered and horizontally if a clutter item is encountered.

As illustrated by the near-vertical red line, all of the UXO, regardless of munitions type, were correctly classified as UXO. In other words, no UXO were mistakenly classified as non-UXO. At the analysts’ recommended stop-dig location, which is identified by the blue dot, over 350 of the actual 450 pieces of non-hazardous metal were also correctly classified.

Figure 2. Results for an example classification demonstration. The line colors in the plot identify munitions classifies as UXO in red, clutter in green, and undifferentiated in yellow. As shown here by the near-vertical red line, all of the UXO were correctly classified

Conclusions
Classification results similar to those shown here have been realized by multiple firms at more than 26 sites across the United States during ESTCP’s Live Site Classification program. UXO classification decisions can now be made based on principled physics-based analysis that is transparent and reproducible. The classification technologies, which include hardware and software components, have initiated a new paradigm in the munitions’ response industry.

References
A Andrews and H Nelson, “Final Report – Implementing Advanced Classification on Munitions Response Sites: A Guide to Informed Decision Making for Project Managers, Regulators, and Contractors”, ESTCP, December 2011

SERDP/ESTCP, “Classification Applied to Munitions
Response”, https://www.serdp-estcp.org/FeaturedInitiatives/Munitions-Response-Initiatives/ClassificationApplied-to-Munitions-Response, 2016