Improved Detection Performance Using A Bayesian Decisiontheoretic Approach

The fundamental goal of target identification using geophysical methods is to achieve a<br>high detection rate along with a low false alarm rate. While many electromagnetic (EM)<br>and magnetic detectors achieve the first of these goals, it is often at the cost of a<br>prohibitively high false alarm rate. In this paper, we present a Bayesian decisiontheoretic<br>approach to target identification using EM data. This approach provides both an<br>improved detection scheme and performance evaluation in the form of ROC curves<br>plotting probability of detection versus false alarm rate. In addition to detailing the<br>Bayesian-based approach, we present selected case studies that utilize broadband<br>electromagnetic data acquired with the GEM-3 sensor. We compare the survey results<br>obtained with standard thresholding analyses to survey results obtained using statistically<br>based signal detection theory. Our preliminary results indicate that the detector derived<br>using signal detection theory is superior to the standard detection schemes that utilize<br>contour plots and thresholding. The results also indicate that detection performance is<br>dramatically improved when multi-frequency data is utilized. Furthermore, since<br>detection performance is observed to be a function of frequency, careful selection of the<br>EM frequencies at which data is recorded may result in further improvements.