Hyperbolically-warped cepstral coefficients for improved micro-Doppler classification

Abstract

Mel-frequency cepstrum coefficients (MFCC) have been used in many recent works as features for micro-Doppler classification. Originally proposed as features for speech recognition, the filter bank applied as part of the computation of the MFCC is designed with spacing according to the mel-frequency scale - a scale based upon the auditory properties of the human ear. However, the frequency composition of micro-Doppler signatures is completely unrelated to the mel-frequency scale. In this work, an alternative set of features computed using a filter bank based on a hyperbolically-warped frequency scale is proposed. A 21.25% increase in the correct classification rate of running, walking, creeping, and crawling is obtained when the proposed hyperbolically-warped cepstral coefficients (HWCC) are used as opposed to MFCC.