Fusion of Data From Forward-Looking Demining Sensors

We present fusion algorithms and detection performance for two sensors of buried land mines capable of forwardlooking operation. The Þrst is a laser Doppler vibrometer (LDV) sensor of acoustically induced soil vibrations developed by the University of Mississippi. The second is a wideband, forward-looking, synthetic aperture groundpenetrating radar developed by SRI. We demonstrate the performance of the individual sensors and fusion algorithms over a data set acquired at Yuma Proving Grounds, AZ. During that collection the LDV sensor was operating in a down-looking mode, but its ability to collect forward-looking data has been conÞrmed in other tests. The amount of fusible data available for these sensors is small, which limits the sophistication and performance of fusion algorithms used with these data. We observed that the GPR and LDV sensors are complementary to a degree. The GPR sensor is signiÞcantly better at detecting metal mines, while the LDV sensor is somewhat better at detecting plastic mines. A classical decision-level “OR” fusion algorithm is shown to be effective for this sensor suite.