Support induced unsupervised weighted decision fusion for wireless sensor networks

This paper present a support induced unsupervised weighted decision fusion method for wireless sensor net works working under unattended environment. A support between two sensors is defined as the possibility of their making the same decision. Mathematically, the support can be formulated as a support relationship matrix, which can be obtained by calculating pairwise support. Conclusion is derived that the principal eigenvector of support relation ship matrix measures the inherent reliability of each sensor. Using this vector as weight of decisions from sensors, an improved decision result can be received. Numerical experiments on both simulated data and data collected by wire less sensor networks from a physical context show that this method result in more superior and robust decision fusion than majority voting rule.