Abstract-We present a novel two-stage methodology for locating a Chemical, Biological, Radiological, or Nuclear (CBRN) source in an urban area using a network of sensors. In contrast to earlier work, our approach does not solve an inverse dispersion problem but relies on data obtained from a simulation of the CBRN dispersion to obtain probabilistic descriptors of sensor measurements under a variety of CBRN release scenarios. At its first stage, subsequent sensor observations under nominal, CBRN event-free conditions are assumed to be independent and identically distributed and we rely on the method of types to detect a CBRN event. Conditional on such an event, subsequent sensor observations are assumed to follow a Markov process. Using composite hypothesis testing we map sensor measurements to a source location chosen out of a discrete set of possible locations. We leverage large deviation techniques to obtain a bound on the localization probability of error and propose several methodologies for fusing sensor data to arrive at a localization decision, including a distributed one. We also address the problem of optimally placing sensors to minimize the localization probability of error. Our techniques are validated numerically using two different CBRN release simulators.

No abstract

Background: We present a Support Vector Machine (SVM) approach to the localization of hazardous particulate releases in an urban area using features constructed only from measurements obtained from a network of sensors. Results: We find high levels of localization accuracy when a reasonable number of noisy sensors are deployed within the environment. We also compare SVM source localization performance to an existing stochastic localization technique over varying degrees of sensor noise and find it favorable for areas prone to urban canyon turbulence effects.Conclusions: This approach is in contrast to earlier works which either use solutions to inverse dispersion problems for localization or apply maximum likelihood techniques. By using established SVM results, we also tackle the problems of release detection and optimal sensor placement.

No abstract