Distributed detection in Neural Network based multihop Wireless Sensor Network

In this paper, a Neural Network based data aggregation approach to detect the binary events in a multi-hop Wireless Sensor Network has been proposed. We envision every node in a network as a unit of neuron which gets trained by using the neural network based back propagation algorithm. As compared to the LMS based Adaptive Weighted Aggregation scheme for tree network, the proposed Neural Network based wireless sensor network approach leads to a significant improvement in detection accuracy without much energy losses due to communication and computation overhead. We also compare the detection accuracy of the proposed Neural Network based scheme with that of the non-adaptive Bayesian approach which requires apriori knowledge of the sensor´s performance indices.