Lightweight target classification for wireless multimedia sensor networks

Abstract

Visual target classification is one of the challenging tasks in resource-constrained wireless sensor networks. This article presents binary and multicast animal classification techniques, which use rule-based decision tree, for wireless multimedia sensor networks. In order to reduce the computational complexity on the sensor nodes, the expensive training phase is carried out by a high-power base station. Then, the best IF-THEN rules are extracted from the decision tree classifier and stored in the sensor nodes before being deployed. This would decrease the learning phase time and the energy consumption, while attaining high classification accuracy. Experimental results demonstrated that the proposed classification model can effectively perform visual target classification in wireless multimedia sensor networks.