Signal - Processing and Sensor Fusion Methods ( Paper

This appendix focuses on the impact of signal-processing techniques on the landmine detection problem and suggests research investments that will allow continued performance improvement. The focus is primarily on processing of electromagnetic induction (EMI) data, although results for other sensors as well as sensor fusion will be discussed. Signal-processing algorithms for landmine detection must detect the presence of an object in the geological background and discriminate signals associated with landmines from signals associated with discrete clutter objects. In general, signal-processing algorithms perform best when the physics that define the problem are integrated within the mathematical constructs underlying the theory of signal processing and pattern recognition. Utilizing experimental data measured under realistic conditions to test the performance of algorithms, and using insight from the data to guide the algorithm development process, has proven to be crucial for reducing false alarm rates in the landmine detection problem. The utilization of computational models describing sensor phenomenology, physics-based feature selection, statistical models of mines and clutter, and spatial information have all led to dramatic reductions in the false alarm rates of landmine detection systems. Because the physics that governs each sensor modality differs, feature sets extracted from data collected by different sensors usually are not consistent across sensors. However, several common approaches to processing the raw