The purpose of this work is to study the feature extraction algorithm of moving vehicles based on signal processing methods. Empirical Mode Decomposition (EMD) is a decomposition algorithm which is used to analyze nonlinear and time-varying signal. Different from the traditional signal analysis method, the decomposition is data-driven and self-adaptive. A feature extraction algorithm based on Empirical Mode Decomposition (EMD) for vehicle classification is studied in this paper. For a long time, feature vectors of ground targets have been extracted by assuming signal short-steady, which acquired short-state features of the target signal. An improved algorithm of energy ratio coefficients extraction based on EMD is proposed in this paper, which is suitable to analyze nonlinear and non-stationary signals. The validity of the algorithm is proved by a test adopting BP network with data sampled from three types of vehicles. Results manifest that the algorithm can accurately characterize the band-energy feature of acoustic targets using the dynamic frequency after EMD decomposition, and get more information for recognition system.