On performance of individual, collective and network detection of propagative sources

We consider the problem of detecting a propagative source amidst background levels based on passive sensor measurements. The source intensity decays with distance according to a known functional form, and the sensor measurements are random with parameters determined by the intensity at the sensor location and the capture area of the sensor. It is generally expected that a collection of sensors offers a better detection performance compared to an individual sensor due to the larger “aggregated” sensor capture area. We show further improvements in the detection performance when the sensors are networked to communicate their measurements and location information to a fusion center to support efficient source localization. If the source can be effectively localized, we establish a superior detection performance of a network over the cases of individual and collective sensors, for a class of detection methods based on probability ratio tests. Our method is based on first localizing the source, and then utilizing an adaptive probability ratio test that uses the estimated source parameters to infer detection. Under fairly general conditions on the source intensity decay functions and underlying measurement distributions, we analytically quantify the performance improvements of the network detection over collective co-located sensors and individual sensors, using the packing number of the underlying state space. We show that these conditions are satisfied in the case of a radiation source, and present simulation and experimental results that illustrate the superior performance of the network detection.