Target localization problem, in which the location of an unknown target is estimated, is one of the key issues in applications of wireless sensor networks (WSNs). Target localization methods that directly use raw sensor data can suffer from uncertainty or disturbance caused by the surrounding environmental elements and noise. Especially, when using WSNs, the limited communication capacity can impose a significant limit on the amount of data that can be processed. Considering these issues, various methods have been proposed, especially using machine learning techniques such as neural networks or support vector machines. In this paper, we employ a multi-class classification algorithm for target localization, in which a pseudo probability map is constructed using modified support vector domain description. A local classification strategy which uses information from local neighbors only is proposed in order to reduce communication costs. Experimental results using an acoustic WSN are compared with Platt’s method to validate the multi-class classification algorithm.
Wireless sensor networks Target localization Support vector learning Multi-class classification