A Modified peripheral auditory model for vehicle classification

Abstract

Peripheral auditory model is a promising method for target classification to extract feature from target acoustic signals. But the general cochlear filter bank is not appropriate for the vehicle acoustic signal because of the obvious difference between the vehicle signal and speech. In this paper, a new method to design the cochlear filter bank for vehicle classification was proposed on the principle of class separability of the patterns. On the basis of the modified peripheral auditory model, monaural model based cepstrum coefficient (MoMCC) was proposed as feature and applied to multi-class vehicle classification. Experimental results show that MoMCC improves both the classification rate and the robustness of the classifier, especially at low SNR.