In the past decade, significant effort has been devoted to the processing of time- and frequency-domain electromagnetic induction (EMI) data for detection and classification of buried unexploded ordnance (UXO). Many of the methods employ a physical model for the scattering of low-frequency electromagnetic fields from UXO-type objects within a statistical signal processing framework. These statistical methods provide a solid foundation for using estimation techniques to determine parameters in the models and hypothesis testing methods for detection and discrimination. While a number of physics-based statistical approaches to the discrimination problem have been proposed, none adequately addresses the issues of uncertain object position and orientation or the processing of spatial information to the extent believed possible.
The objective of this project was to better understand the utility of time- and frequency-domain EMI data for UXO discrimination and to develop algorithms that optimally use both classes of information when the position and orientation of the object are uncertain. The following issues were addressed: (1) How accurately can object orientation and position be estimated and how sensitive are discrimination results to these estimates? (2) How does the addition of spatial information quantitatively improve discrimination performance for both time- and frequency-domain EMI systems? (3) When and to what extent is time- and frequency-domain EMI fusion useful?
A two-stage approach to the problem outlined above was undertaken. First, physically relevant parameters were extracted from the data. Second, these estimates were used in a statistically based discrimination stage to separate UXO from clutter. Underlying this effort was a dipole scattering model linking the input fields, the object, and the observed data. The model was explicit in the scattering-based features to be used for discrimination, the (x, y, z) position of the object, and its orientation relative to the sensor, thus allowing uncertainty in object location and orientation to be directly addressed. The common physical modeling structure underlying the time- and frequency-domain algorithms naturally allows for sensor fusion.
Physics-based signal processing methods were developed to distinguish clutter objects from UXO and to classify the type of ordnance. In addition, computationally tractable techniques were developed to account for uncertainties in the sensor’s position and orientation. Preliminary results using synthetic data indicated the viability of the prototype algorithms.Follow-on research is being conducted under SERDP project MR-1379 to refine and enhance the signal processing algorithms, develop a library of UXO time- and frequency-domain signatures, and validate the algorithms using laboratory training and field data.
This project has led to the development, analysis, and validation of a general purpose approach to the EMI sensing classification problem, which explicitly accounts for uncertainties in object position and orientation. Such an approach, ultimately, will lead to improvements in UXO discrimination. (SEED Project Completed – 2003)