Distributed sensing for quickest change detection of point radiation sources

We consider the problem of distributed detection of a radioactive source using a network of emission count sensors. Sensor nodes observe their environment and a central fusion node attempts to detect a change in the joint probability distribution due to the appearance of a hazardous source at an unknown time and location. We consider a minimax-type distributed change-point detection problem that minimizes detection delay for a desired false alarm rate. A statistical model of the radiation source detection problem is formulated where sensors observations are correlated with non-identical distributions. We first derive a centralized detection algorithm that is asymptotically optimal for vanishing false alarm rate. Then we analyze the performance loss, as measured by the detection latency, when sensor counts are quantized at each sensor node. The detection latency of the centralized rule provides a lower bound on performance for the proposed distributed method. The empirical results indicate that the distributed detection strategy provides a reasonable tradeoff between latency and information bandwidth.