Adaptive MaxEnt modeling of distributed decision fusion without knowledge of prior probabilities of local decisions

We introduce the method of the Maximum Entropy (MaxEnt) model for fusing local decisions in a distributed multiple sensor system. The fusion center receives local binary decisions in the usual parallel architecture. No assumptions are made about knowing any local decision rules. Our approach is based on the concept of machine learning, wherein the MaxEnt parametric model is used for supervised classification and prediction serving as the central (global) decision rule. Therefore, the system is able to learn the detection performance of the sensors as a function of time without prior knowledge of the actual probabilities of local decisions, only requiring an initial set of random training data. Thus it is demonstrated that the system is adaptive and can learn contextual changes of the sensors. Furthermore, we provide simulation results comparing the MaxEnt fusion center performance with published results using both the Bayesian formulation and Neyman-Pearson criterion and with MaxEnt achieving the best, realistic detection performance demonstrating the effectiveness of the method.