An Improvement on Discrete Wavelet Transform-based Algorithm for Vehicle Classification in Wireless Sensor Networks

In this paper, we present an improvement on the discrete wavelet transform (DWT)-based feature extraction algorithm used in vehicle classification in wireless sensor networks (WSN) by providing a rule of selecting the best number of resolution levels which will improve the classification rate and reduce the energy consumption in local computation. DWT can provide time-frequency multi-resolution analysis (MRA) which is fast and can greatly reduce the dimensions of feature vectors. We found out that, when using wavelet, the number of resolution levels will greatly influence the effect of feature extraction and energy consumption. In this paper, different numbers of resolution levels and corresponding time of computation are discussed. And a rule of selecting the best number of resolution levels is given. To test our algorithm, acoustic signals emitted by two kinds of vehicles are investigated, and a k-nearest-neighbor method is used as classifier. The experiment shows that the rule provided can obviously improve the classification rate