Detection and Identification of Vehicles Based on Their Spark-Free Unintended Electromagnetic Emissions

Detection and identification of devices using their electromagnetic emissions is a widespread practice. Vehicles are among the most complex with emissions from a whole range of electrical and mechanical components. The authors in a previous work [11], detected and identified vehicles using their electromagnetic emissions by neural network analysis of data derived from the fast Fourier transform (FFT) of measurements. The method was successful provided there was an ignition spark event captured. In this letter, the authors focus on the no-spark case and instead of FFT, use wavelet packet analysis (WPA). WPA, by providing arbitrary time-frequency resolution, enables analyzing signals of stationary and nonstationary nature. It has better time representation than Fourier analysis and better high frequency resolution than Wavelet analysis. WPA subimages are further analyzed to obtain feature vectors of log energy entropy. Similar to the previous work [11], training and testing is done on separate days. Emissions from three cars and ambient noise are analyzed and then classified using a multilayer perceptron. 100% detection and identification rate is accomplished when there is no ignition spark event present.