Adaptive fusion by reinforcement learning for distributed detection systems

Chair and Varshney (1986) have derived an optimal rule for fusing decisions based on the Bayeslan criterion. To implement the rule, the probability of detection P _D and the probability of false alarm P _F for each detector must be known, but this information is not always available in practice. An adaptive fusion model which estimates the P _D and P _F adaptively by a simple counting process is presented. Since reference signals are not given the decision of a local detector is arbitrated by the fused decision of all the other local detectors. Furthermore, the fused results of the other local decisions are classified as "reliable" and "unreliable". Only reliable decisions are used to develop the rule. Analysis on classifying the fused decisions in term of reducing the estimation error is given, and simulation results which conform to our analysis are presented.