Applying system identification and neural networks to the efficient discrimination of unexploded ordnance

This paper describes system identification and neural network methods which, when applied to simulated noisy radar returns from unexploded ordnance (UXO), result in significantly better performance than more traditional methods. The methods are applicable to wideband ground penetrating radar (GPR) waveforms. The target set consists of a conducting cone and cylinder of similar dimensions. The classification is based on features derived by the singularity expansion method.