Vehicle Type Recognition Based on Increasing Hierarchical Feature Convolutional Neural Networks

To improve the accuracy and reduce convergence rate for vehicle type recognition, this paper proposes a novel method based on IHFCNN (Increasing Hierarchical Feature Convolutional Neural Networks). Firstly, utilizing the characteristics of biological vision, a new network layer based on classical convolutional neural network is presented to simulate the hierarchical processing of information. Secondly, features are extracted from different layers, and then a fully-connected layer is connected. Finally, center loss is introduced to output layer, and Softmax classifier is applied for vehicle type recognition. The experimental results show that, compared with the classic networks as AlexNet, GoogLeNet and VGG16, the proposed method reduces the number of convergent iterations significantly and improves the recognition accuracy up to 97.78% in the database with 321678 training images and 8490 for testing in real application scenarios.