Abstract: Aiming at the defects of the current recognition methods using single signal, we propose a vehicle recognition method based on the collaboration representation of acoustical and vibrative signals. Firstly, Mel-frequency cepstral coefficients (MFCCs) are used to extract the acoustical and vibrative features of vehicles. Then, multitask training of classification is carried out separately by two kinds of signal features. Finally, the reconstruction error of multitask collaboration representation is obtained through the signal features and the target is classified according to the reconstruction error. Experiments indicate that this method has better classification effect and higher recognition efficiency, compared with the existing methods.

Key words: vehicle recognition, collaboration representation, multitask classification, feature extraction, reconstruction error