Multisensor Estimation And Decision Fusion

Multi-sensor data fusion has two main classes: estimation fusion, decision fusion. In estimation fusion, three fusion architectures are considered: centralized, distributed, and hybrid. A unified linear model and general framework for these three architectures was established before optimal centralized and distributed fusion rules have been presented. This dissertation is mainly concerned with the performance of distributed and centralized fusion for best linear unbiased estimation (BLUE), and optimal weighted least squares (WLS) estimation. Necessary and sufficient condition for optimal distributed fusion rules to have identical performance as their centralized counterparts is presented. Further, measures of relative efficiency of distributed fusion compared with centralized fusion are proposed.In decision .fusion problem, this dissertation mainly discussed the multi-sensor Neyman-Pearson type sequential decision with correlated sensor observations, and achieved the optimal sensor rule under given fusion rule. It was developed from the new result of centralized Neyman-Pearson type sequential decision. The new decision method has better performance than the traditional Wald's sequential decision in two folds: more sufficient taking advantage of the two permitted error probabilities; guaranteeing to have an upper bound w. p. 1 of the number of observations required for terminating the decision. Simulations in this dissertation provide additional supports to the above.