Neural network approaches and MSPCA in vehicle acoustic signal classification using wireless sensor networks

Abstract

Acoustic communication has been widely used in wireless sensor networks. Vehicle acoustic signals have long been considered as unwanted traffic noise. In this research acoustic signals generated by each vehicle will be used to detect its presence and classify the type. The goal of multiscale PCA (MSPCA) is to reconstruct a simplified multivariate signal, starting from a multivariate signal and using a simple representation at each resolution level. Multiscale principal components analysis generalizes the PCA of a multivariate signal represented as a matrix by simultaneously performing a PCA on the matrices of details at different levels. By selecting the numbers of retained principal components, simplified signals can be reconstructed. These simplified signals are used for extracting the features. Six different features of the vehicle acoustic signals are calculated for the pre-processed acoustic vehicle signals and then further utilized as input to the classification system. These features include Signal Energy, Energy Entropy, Zero-Crossing Rate, Spectral Roll-Off, Spectral Centroid and Spectral Flux. Acoustic signal classification consists of extracting the features from a sound, and of using these features to identify classes the sound is liable to fit. Neural network approaches used here are KNN, PNN and BPN and these three approaches are combined with the MSPCA to obtain better accuracy.