Triaxial Modeling and Target Classification of Multichannel, Multicomponent EM Data for UXO Discrimination

Abstract

Time-domain electromagnetic data collected for discrimination of unexploded ordnance (UXO) span one spatial component in one time gate to three spatial components and multiple time channels. The relative merits of additional temporal and spatial-component information were assessed by analyzing decimated data from the Geonics EM61-3D. The responses of 25 unique ordnance- 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. Performance was measured by the fraction of false positives for ordnance-like objects at 91% true positives. The best false-positive rate for the full multicomponent, multichannel data was 5%, which was achieved using the power-law slope of the time decay of the secondary-field EMF: ordnance-like items remain in “intermediate” time longer and display a characteristic [Formula: see text] slope, whereas scrap-like items (even those made of steel) transition to “late” time and the recovered power-law indices vary widely. However, this excellent performance is largely due to the fact that the ordnance-like objects are often larger than the scrap-like objects in this data set. Better generalization may be obtained with discriminants based solely on shape—ordnance-like objects are usually axisymmetric about a long axis—which yielded 32% false positives, again in the power-law slope. In contrast, the best false-positive rate for single-component, single-channel data was 68%. Other instrument configurations (single-component, multichannel or multicomponent, single channel) were generally ranked intermediately between these extremes. Additional experiments with intermediate numbers and positions of time gates revealed that the best discrimination is achieved when recording a broad time range for the secondary-field decay [Formula: see text]; the number or density of time gates has a smaller impact. These results demonstrate in an internally consistent way the significant performance improvement in UXO discrimination that can be achieved with multicomponent, multichannel electromagnetic sensors.