In the recent years, a wireless sensor network (WSN) has become one of key technologies to enable people``s ubiquitous computing life. Advances in WSNs make many of impossible possible, and the various needs of WSN applications are emerging increasingly. Target classification in WSNs is one of the most important and demanding technologies used to meet such various needs. However, considerable complicated processing is required for target classification, which is opposed to the resource limitation of sensor nodes.
The objective of this thesis is to develop an efficient target classification scheme for WSNs which is called PATaCS. The proposed PATaCS consists of target detection, target classification, and data fusion for processing efficiency. Specifically, since each processing component in the PATaCS is closely related with one another, they are required to be explored together for optimization. In this thesis, the PATaCS scheme is proposed, discussed, and evaluated by the classification accuracy of vehicles with acoustic and seismic signals collected from a real sensor field. In addition, some implementations are given to show their feasibility. Each component of the PATaCS scheme is described as follows:
First, target detection should be done in advance of target classification. Among many detection theories, the constant false alarm rate (CFAR) algorithm is usually employed since the environmental characteristics of WSN are nonstationary and unpredictable. Accordingly, we use a sort of the CFAR algorithms to detect a target in our works.
Second, when a classification algorithm for WSN nodes is designed, a parametric approach should be more preferred to non-parametric one due to the hard limitation in resources. As a parametric classifier, the Hidden Markov Model (HMM) algorithm or the Gaussian mixture model (GMM) algorithm does not only show good performances for target classification in WSNs but also require very small resources, making it suitable f...