Improved signal processing approaches for landmine detection

In our previous work, we have shown that a model-based Bayesian approach to the detection of mines affords significant performance gains for both time-domain and frequency-domain electromagnetic induction (EMI) sensor over standard thresholding techniques(1,2). Our methodology merges physical models of the evoked target response with a probabilistic description of the clutter, and provides both an optimal detection algorithm, under a specific set of assumptions, and performance evaluation measures in the form of probability of detection (Pd) and probability of false alarm (Pf,). This approach was validated on data obtained during the DARPA Backgrounds Clutter Data Collection Experiment(3). In this paper, we review these results, and also present theoretical results that show that utilization of the entire multi-channel time-domain or frequency-domain waveform always affords an improvement in detection performance over single-channel systems. In addition, we present results that indicate that including spatial information into the processor substantially reduces the false alarm rates over processors in which this spatial information is not included. Finally, we discuss the performance of a more advanced detector, also formulated using signal detection theory, in which uncertainty in the placement of the mine in the environment is incorporated into the detector.