EVIDENTIAL REASONING APPROACH TO DISTRIBUTED MULTIPLE-HYPOTHESIS DETECTION.

A distributed detection system consisting of n local detectors and a fusion center is considered. The local detectors send their inferences with regard to a set of M candidate hypotheses to the fusion center where the final decision is made. Unlike conventional detectors which produce single-hypothesis hard decisions, each local detector produces a 'subset' of the M hypotheses and a discrete confidence level associated with that subset. The fusion center then employs A. P. Dempster's (1986) combining rule for evidence aggregation. The optimum confidence-based partitioning of local decision space with respect to the system performance is studied. For the M equals 3 case, it is shown that the presented system greatly outperforms the one in which each local detector provides a single-hypothesis hard decision.