Model-based approaches to unexploded ordnance (UXO) discrimination seek to recover physical properties of the target itself (size, shape, orientation, position) and therefore are most useful for target classification. However, strong trade-offs exist between model accuracy and efficiency.
The primary objective of this project was to compare three quantitative approaches to modeling electromagnetic induction for UXO discrimination—a phenomenological dipole model, a semianalytic theory, and a finite-element numerical method. The secondary objective was to quantify the value of multiple spatial components and time channels.
Phenomenological dipole models, a popular semi-empirical method, treat targets as a group of infinitesimal, orthogonal dipoles and construct responses for arbitrary position and orientation through linear superposition. Classification can be improved by experimentally determining the directional responses of targets of interest. The technique is very fast but is limited to distances relatively far away from the source and receiver, assumes no environmental effects, and requires prior data for each object to be classified. A semianalytic theory is relatively fast, models the full field at arbitrary distances, and can include ground conductivity but restricts target shapes to solid triaxial ellipsoids. Numerical models such as the finite-element method offer the most accurate solutions for arbitrary objects and environments but are slow, unsuited to parameter estimation, and require that responses from all potential targets be cataloged.
Of the three physics-based models investigated, only the simplest—the dipole—is presently suitable for UXO studies. Researchers extended this model to a full triaxial, time-dependent representation. This model was used to infer target properties using multicomponent, multichannel time-domain electromagnetic signatures of 45 seeded objects obtained using the Geonics EM61-3D. Seeded targets were divided into “ordnance-like” and “scrap-like;” the former were axisymmetric about a long axis whereas the latter were not. The triaxial time-dependent dipole model was used to estimate up to 18 parameters regarding target size, shape, position, and orientation. When nondimensionalized by the values along the object’s apparent longest axis, eight parameters are relevant to target size and shape. The general, time-dependent triaxial dipole model was successful in discrimination tests. The responses of 25 unique ordnance-like and scrap-like objects buried in 45 depth-orientation states were modeled as triaxial time-dependent dipoles and were classified using the properties of the inferred model parameters. This work demonstrated the high performance in UXO discrimination that can be achieved with multicomponent, multichannel electromagnetic sensors, as well as the value of relatively simple modeling and discrimination procedures.
This project identified models that are likely to be successful in real-world UXO surveys where target reports are often expected within a day. The project also points to optimum sensor designs. (SEED Project Completed – 2002)