Distributed and Efficient Classifiers for Wireless Audio-Sensor Networks
Technical report TR08-02. We consider the problem of vehicle classification using acoustic signals captured within a sensor network. The sensors perform collaborative decision and/or data fusion in a distributed and energy efficient manner. We present a distributed cluster-based algorithm, where sensors form clusters on-demand for the sake of running the classification task. We aim at minimizing the energy costs incurred due to the transmission of the feature vectors among collaborating sensors within a cluster. To this end, we present schemes to generate effective feature vectors of low dimension. An experimental study has been conducted using real acoustic signals of military vehicles recorded during DARPA's Sensit/IXOs project. The features generated through our proposed schemes are evaluated using K-Nearest Neighbor (k-NN) and Maximum Likelihood (ML) classifiers. Performance results indicate that the proposed schemes are effective in terms of classification accuracy, and can even outperform previously proposed approaches, but, in addition, they are also efficient in terms of communication overhead. | TRID-ID TR08-02