Toward Performance Optimization With Data Fusion In Wireless Sensor Detection Networks

Wireless sensor networks for Mission-critical detection often face the fundamentalchallenge of meeting stringent performance requirements imposed by users, such as the lowfalse alarm rate and the high detection probability. However, the quality of the data sampledby sensors is quite different since the physical environment is stochastic and, the devicequality is different between sensors. As an effective method to enhance system accuracywhich can eliminating the effect of noise and other factors, data fusion scheme collects andfuses data from different sensors, and adopting collaboration between them, it can get bettersystem performance in low data quality. Nevertheless, the traditional data fusion methodsusually adopt simple Vote based mechanism and Bayes criteria which need the probability andcost factor of the events as a prior. However, the estimation of those parameters is very hardaccording to the current research results, which is one of the most challenges that has becomea bottleneck in practical applications.In this thesis, the probability data fusion model is proposed, which use N-P and Minmaxcriterion respectively. We have designed and analyzed the decision fusion mechanism foroptimizing the performance. Firstly, based on the traditional decision fusion, the detectionscheme has been designed which utilizes the N-P criteria such that the maximum detectionprobability can be achieved while ensuring the preset false alarm rate constraint. We have alsodeducted the optimal local threshold and global threshold, respectively. Secondly, we proposea decision fusion scheme according to the minmax criterion, and analyzed the optimal fusionthreshold without the priori event occurrence probability. Finally, numerical analysis andsimulations are conducted at different simulation scenarios, and both of them have verifiedthe effectiveness and efficiency of the proposed scheme.