Government programs are developing multi-sensor detection platforms that can provide multiple high-quality signals with which to classify subsurface objects. It is estimated that 11 million acres of military land need assessment to identify subsurface unexploded ordnance (UXO), with costs estimated to be about $1.4M per acre. The ranges where UXO can be found are distributed throughout the country, where clutter conditions vary significantly. It is the clutter, which causes a high false alarm rate in UXO detection, that drives up the cost of remediation. The hypothesis of this project is that by fusing multi-sensor data with novel statistical signal processing techniques, the rate of false alarm in UXO detection can be reduced dramatically.

This project will investigate the phenomenological aspects of the UXO detection, location, and discrimination problem using electro magnetic induction (EMI), radar, seismic, and magnetometer sensors. The fundamental insight garnered by characterizing the underlying physics will be transitioned into high-performance sensor fusion and signal-processing algorithms for enhanced detection, location, and discrimination of buried UXO under a range of environmental conditions.

The technical approach will employ synergistic research activities in modeling, signal processing, and sensor fusion. The researchers will perform phenomenological modeling of wave propagation and scattering for ultra-wideband radar, seismic, and EMI sensors. The phenomenological studies will be performed in collaboration with SERDP-supported sensor-development programs under way at the Naval Research Laboratory, the Army Research Laboratory, and BBN. The previously developed models will be extended to allow arbitrary numbers of soil layers and arbitrary target shape and orientation and to account for all interactions accurately. The use of these models will quantify the target types, depths, and soil conditions for which radar is an appropriate sensor.

Modeling work to date has focused on the following three sensors: magnetometer, time- and frequency-domain EMI, and Boom-synthetic aperture radar (SAR). Team members have applied a simple dipole model for magnetometer data to field data and have successfully used it to estimate target depth. The project currently is investigating the utility of other parameters estimated with the model as target discriminants. The team has developed a full method of moment model for predicting the fields recorded by EMI sensors as well as a simpler model that assumes that the induced magnetic fields can be characterized by a dipole. The project also is investigating the performance and limitations of the simpler model. For the Boom-SAR, team members have developed a multi-level fast-multipole algorithm for modeling radar scattering. The model results in considerable computational savings and has been compared favorably to measured radar data.

The goal of this project is to develop algorithms that substantially reduce the false alarm rates associated with individual sensors and optimally combine information across sensors to further reduce the false alarm rate.