This paper proposes a cepstrum coefficient model to overcome deficiency in the classic Mel-frequency cepstral coefficients (MFCC). Based on the human auditory model, two improvements are made in computing the new cepstrum features for acoustic target recognition. First, by allowing variable time-frequency resolution and to be immune to noise, wavelet packet transform (WPT) is used instead of FFT in the computation of classic MFCC. This overcomes deficiency of single resolution and reduces sensitivity to complicated battlefield environments. Second, based on the human auditory model, compression with the exponential-law is used instead of the fixed logarithm amplitude spectrum used in the classic MFCC. Experimental results show that the new algorithm can achieve significant advancement over the former method in accuracy and robustness.