ASSESSING GEOPHYSICAL LANDMINE DETECTION METHODS IN SAND AND SOIL BEDS SIMILAR TO DONETSK REGION OF EASTERN UKRAINE

Landmines have undergone many technological advances since the World Wars of the 20^th Century. However, the most common detection methods (i.e. prodding and metal detectors) have not significantly changed. Current research is focused on developing a multisensor (holographic radar, impulse radar, metal detection and infrared imaging) semi-automated, robotic device to reduce false alarms, missed mines, and human casualties. This study has a specific focus on evaluating methods that might be useful in the Donetsk conflict zone of Eastern Ukraine. Two test beds were constructed in Lancaster, PA; one to provide ideal conditions (uniform sand) for evaluation of best-case detection, and another with soils resembling the dominant chernozemic soils of the Donetsk region. Each bed was seeded ten times with various random combinations of actual (inert) landmines and clutter, with clutter outnumbering mines by at least 6:1. Plan-view subsurface images were generated by each method, and presented to 37 minimally-trained volunteer subjects who were asked to identify the objects in each image, and label them as either mines or clutter. In both the sand and soil beds, more than half of the mines were detected (53% and 60% respectively at p=0.05) by these untrained operators. Interestingly, landmines in the mock Donetsk soil beds were detected more frequently than in sand beds, and large anti-tank mines were significantly more detectable than the smaller anti-personnel mines. The testing results were also used to generate a receiver operator characteristic (ROC) curve relating probability of detection (PD) to probability of false alarm (PFA), with naturally-varying operator judgment/confidence acting as a proxy for instrument sensitivity. Using the multisensor images, these untrained operators produced an ROC curve for Donetsk soils that falls within the range of published curves for individual instruments operated by trained experts. These results indicate that the chosen sensors should be suitable for Donetsk soils (operators rated impulse radar most helpful, and infrared least helpful), and that with fully-trained operators, this sensor combination should be able to match, or even exceed the performance of single-sensor landmine detection systems.