Particle Swarm Optimization Clustering for Target Classification in Wireless Sensor Networks

Abstract

Wireless sensor networks (WSNs) is an emerging technology that enables information retrieval from the environment by densely deployed tiny, low-cost and low-power wireless device called sensor nodes. In this paper, we explore a two-tiered WSN model containing both static and mobile sensor nodes, and focus on vehicular target classification with the small sample kernel classifier of support vector machine (SVM). Clustering is employed to achieve energy efficiency in battery powered WSNs and facilitate collaborative processing that promises to improve classification accuracy. Since clustering is an NP-hard problem, particle swarm optimization (PSO), a stochastic optimization technique emulating the behavior of a flock of birds, is used to search for the optimal cluster formation. In addition, a simple yet effective cluster number estimate technique is put forward, which takes into account the maximum communication range. Collaborative target classification is implemented with a simple voting scheme. Simulation experiments show that PSO clustering is effective and collaborative SVM classification markedly improves target classification accuracy.