Vehicle Type Recognition based on Multi-branch and Multi-Layer Features

To improve the accuracy and reduce the computational complexity of neural networks for vehicle type recognition, this paper proposes a novel method based on Multi branch and Multi-layer features. First of all, each car-face image is divided into multiple sub-images according to texture features'characteristic. Secondly, global and local features are extracted using several convolutional neural networks (CNN) in different layers then connected to a fully-connected layer. Finally, Softmax classifier is used for vehicle type recognition. Experimental results show that valid global and local features in both top and bottom layers are extracted by the proposed method. Furthermore, convergent efficiency and accuracy of recognition are improved.