Incremental distributed identification of Markov random field models in wireless sensor networks
IEEE Transactions on Signal Processing
| Hi-index | 35.69 |
In this paper, the problem of event-region detection and the related problem of boundary-region detection in wireless sensor networks (WSNs) are explored. Decentralized detection techniques are developed to tackle these problems. In contrast to the standard decentralized detection where all sensors observe a homogeneous phenomenon, sensor nodes considered here could observe heterogeneous regions, and the underlying phenomena at sensor nodes can be correlated. Each node makes its decisions regarding which region it locates and/or if it is a boundary sensor. In particular, a space-memory fusion rule that considers both space-memory information and local detection performance is derived for event-region detection. The proposed fusion rule performs in a sequential manner. Communication constraints and possible sensor faults, which are important issues in WSNs, are integrated in the design of the space-memory fusion rule. Specifically, a low-communication-rate sensor fusion approach is derived by incorporating the communication cost in either a Bayesian risk or a constrained optimization formulation. The a priori sensor-fault model is adopted to address the issue of fault-tolerance capability. The problem of boundary-region detection is transformed into boundary-sensor detection, which is then formulated as a binary-hypothesis-testing problem. Simulation results demonstrate the advantages of the proposed fusion rule in both event-region and boundary-region detections.