The Mel-Frequency Cepstral Coefficients(MFCC) are susceptible to noise in field vehicle recognition. To enhance the robustness of features, this paper proposed a weighted and adaptive feature extraction algorithm based on the MFCC method. Firstly, the energy to entropy ratio method was used to detect the endpoint of field vehicle's acoustic signal. Then, the fundamental frequency of vehicle's acoustic signal was extracted. The triangular filter bank was adaptively constructed according to the fundamental frequency so as to improve the filter’s sensitivity to the fundamental frequency. Finally, the obtained frequency was weighted with fisher’s ratio. Compared with the traditional MFCCs, the experimental results show that the improved MFCCs improve the recognition accuracy by 7.10%, reduce the false alarm rate by 3.93% and reduce the leakage alarm rate by 7.10% in field vehicle recognition.