Distributed and disposable sensors at ARL and the ASCTA

Classification of Military Ground Vehicles Using Time Domain Harmonics' Amplitudes

“…ANNs derive their computational power from their ability to learn combined with the nonlinearity of neurons. Simple feedforward neural networks (FNNs) are typically used in vehicle classification [10], [15], [16], [27]. A three-layer FNN with sigmoid transfer function is utilized, where the number of neurons in the input layer is equivalent to the length of the characteristic FV, and the number of neurons in the output layer is equivalent to the number of vehicle classes.…”

“…The motivation for using frequency domain features is the energy concentration of the acoustic signal, which lies mostly in the range 20-400 Hz. In [8] and [9], consecutive FFT components were selected for generating elements of the feature vector (FV), while in [13], [15], [16], only harmonically related FFT peaks that exceeded the noise level were selected for generating the FV. Harmonically related peaks result in a lower dimension FV, but determining those peaks has two requirements: first, 0018-9456/$26.00 © 2011 IEEE the effective resolution of the computed FFT components has to be about 1 Hz or less, which raises the overall number of complex multiplications; second, a good estimate of the fundamental frequency (FF) and the number of harmonics should be obtained.…”

“…The TDHA method aims to increase the life time of individual sensor nodes. The TDHA method can also be used in many applications that utilize acoustic/seismic signatures for source identification as in traffic control [20], surveillance [10], [15], [16], health and environmental monitoring, and fault diagnosis [21].…”