Efficient Sensor Selection for Active Information Fusion

In our previous paper, we formalized an active information fusion framework based on dynamic Bayesian networks to provide active information fusion. This paper focuses on a central issue of active information fusion, i.e., the efficient identification of a subset of sensors that are most decision relevant and cost effective. Determining the most informative and cost-effective sensors requires an evaluation of all the possible subsets of sensors, which is computationally intractable, particularly when information-theoretic criterion such as mutual information is used. To overcome this challenge, we propose a new quantitative measure for sensor synergy based on which a sensor synergy graph is constructed. Using the sensor synergy graph, we first introduce an alternative measure to multisensor mutual information for characterizing the sensor information gain. We then propose an approximated nonmyopic sensor selection method that can efficiently and near-optimally select a subset of sensors for active fusion. The simulation study demonstrates both the performance and the efficiency of the proposed sensor selection method.