Spectral Eigen Index: Military vehicle fingerprinting using Eigen analysis in spectral domain

It is of great interest for military to identify specific vehicles using vibrating data collected by sensors. In view of the special spectral properties of military vehicles, for a vibration measurement, we first formulate a matrix by stacking the overlapped short-duration measures, the eigenvector of the largest eigenvalue is taken to represent the original measurement after a sequence of Gaussian smoothing, high frequency suppression, sub-sampling and an additional Fourier transform. A new vibration index, denoted by spectral eigenvector index, is formed as the fingerprint of the vehicle in consequence. A feed-forward Neural Network N with 70 hidden nodes is trained as the vehicle type classifier. Vehicle classification using N trained based on this new index can achieve valuable accuracies for a standard testbed of military vehicles.