In order to improve speed and accuracy of automatic vehicle identification system

a 13-characteristic extraction process is used

and the feature vectors are inputted into network.An Elman recursion neural network is selected which including an undertaking layer as a step delay operator.This network when updating the weight value not only considers the current gradient direction

but also considers the former gradient direction

thereby reducing sensitivity of the network performance parameters and effectively suppressing the emergence of local minima.Finally by comparing with BP network training

the results show in the recognition of speed and accuracy the Elman recurrent neural network has more advantages.