Objective

There are two basic parts to the process of discriminating between buried unexploded ordnance (UXO) and non-hazardous metallic clutter in the ground. First, target parameters or features are extracted from data collected over the target. Then the target parameters are supplied to a set of decision rules that classify the target as UXO or clutter. UXO/clutter discrimination performance to date has not lived up to expectations. There are problems with both parts of the process. First, discrimination techniques rely on inverting spatially mapped electromagnetic induction (EMI) data to estimate target parameters and are sensitive to noise and small errors in data mapping. Second, discrimination performance depends on the training data used in determining the boundaries of the decision regions. Although an extensive data base of UXO signatures has been compiled under SERDP project MR-1313, there is limited clutter data collected under controlled conditions available for training. This project was intended to help rectify this deficiency by collecting and analyzing EMI data for a variety of clutter items.

Project objectives were to establish relationships between the parameters that characterize a target's EMI signature and the physical attributes of the target that can be used to develop effective classification rules; and devise robust processing and analysis procedures for estimating the target parameters from data collected above an unknown target without having to spatially map the data.

Technical Approach

Processing of EMI response data involves fitting the data to simple parametric models. A physics-based two-component model has an algebraic form for the early time response paired with an exponential decay for the late time response. The two components (algebraic and exponential) in the model represent contributions to the response from magnetic surface modes and body modes.

A simple empirical parameterization can be used in a similar manner. This model combines the algebraic and exponential characteristics of the EMI response using a single power law exponent (γ) and an exponential decay time scale (τ). The model is simpler than the complete two-component response model, which includes modal amplitude and bandwidth parameters.

Results

The researchers determined that the simple empirical parameterization can be used to adequately represent the response for all but the earliest part of the eddy current decay. As might be anticipated from the two-component model results, the researchers found that distributions of the power law and exponential decay parameter values for several classes of clutter items are varied from those of different munitions items. The trajectories of these parameter values that are swept out as a sensor is moved about over a munitions item are distinctive, and the researchers devised a simple procedure for estimating the shapes of the principal axis polarizability curves for munitions items from data collected as a sensor is blindly moved about over the item.

The time domain data measures only the eddy current decay component of the EMI response of an object. However, there is another part of the response distinct from the eddy current response. Steel objects (like most munitions and clutter) become magnetized while the primary field is on. The researchers analyzed this bulk magnetization component of the response for munitions and clutter items using frequency domain data. An object's shape affects how it responds to magnetic fields. The effects are represented by demagnetizing factors that depend on the direction of the applied field relative to the object and are determined by the physical details of how the object responds to externally applied magnetic fields. These demagnetizing factors are another amplitude-independent response parameter and can be determined by fitting the in-phase part of the frequency domain response with a simple model. Distributions of the demagnetizing factors for munitions and clutter items are distinctly different. This reflects real shape differences. The munitions items are basically 4:1 aspect ratio cylinders as far as the magnetization response is concerned, and their demagnetizing factors cluster accordingly. Values for the clutter items scatter over a much wider range, and the maximum demagnetizing factors for the clutter items are often significantly larger than for the munitions. These are consequences of the facts that the clutter items tend to be irregularly shaped and that many of them have one dimension significantly smaller than the other two.

Benefits

The results of this project can benefit research and development efforts aimed at improving the performance of existing and planned new technology in UXO/clutter discrimination. The robust parameter estimation procedures developed are directly applicable to existing sensor technologies and could be of use in the specification of future sensors. The clutter signature database assembled are available in digital form for distribution.

  • Analysis,

  • Electromagnetic Induction (EMI),

  • Physics-based,

  • Classifiers,