A Feature Extraction Method for Wheeled and Tracked Vehicle Classification Based on Geologic Model

Abstract

Seismic signal is widely used in ground vehicle classification due to its inherent characteristics. But the generalization accuracy of classifier is heavily degraded due to different underlying geologies. To overcome the weakness of the seismic signal, a feature extraction method is proposed in this paper. The extracted feature is the cepstrum of the seismic signal whose logarithmic power spectrum density will be preprocessed to suppress the geology related components, which is based on the special characteristics of the employed geologic model, before further calculations. The efficiency of the proposed feature is verified with a mixed database taking from our field experiments and SensIT project.