Approaches for unexploded ordnance (UXO) discrimination signal processing to date assume particular models for the subsurface objects that produce recorded signals. Optimization calculations then estimate the location and type of these sources. This can be a laborious procedure, even for simple models of typical sources. The models chosen (e.g., tri-axial magnetic dipoles to represent metallic items such as clutter or UXO) are usually of limited validity or applicability. Further, the analyst rarely knows how many objects are contributing to the signal and how important each is.

The objective of this project was to explore ways to process magnetometer or electromagnetic induction (EMI) sensor data so that UXO discrimination processing can be enhanced, focusing on the suppression of clutter.

Technical Approach

This project analyzed aboveground signals to see where they lead below ground or in extensions of the data above ground into less cluttered mixes of the signals. Based on physically rigorous governing equations, the computations treat the ground surface data as an extended boundary condition. On this basis, the algorithms solve numerically for a complete magnetic field or source picture over regions above or below ground. The approach requires magnetic field data over the locale of an anomaly corresponding to a single excitation field, so that the data over a measurement area can be used together as a consistent boundary condition.


Innovative signal processing techniques, derived from basic physics, were applied to EMI data to enhance its suitability for UXO discrimination processing. In all simulations and field tests, upward continuation (UC) of surface data succeeded in suppressing clutter signals relative to those from deeper UXOs. This was the case even when the clutter signal was two to three times the strength of the broader UXO response in which it was embedded. Downward continuation (DCN) for the purposes of focusing on subsurface source locations is inherently more problematic. Ill-conditioning and attendant amplification of signal noise tends to plague DCN and is likely to occur to some degree even in UC. Methods were identified and developed for analyzing and controlling such ill-conditioning reliably by spectral truncation. The predictive methods require no particular target models, no optimizations, and no searches for target location, orientation, and properties. Altogether, the results of the project form the basis for development of simple and fast "model-free" discrimination algorithms.


Utilizing above and below ground projections of surface data could provide a more complete, continuous picture of subsurface anomalies to support the determination of whether they are munitions or clutter. Applying these methods may lead to improved discrimination, an expanded ability to cost-effectively characterize munitions and explosives of concern (MEC) sites, and increased capabilities for a wide diversity of site conditions.