Vehicle classification based on the fusion of deep network features and traditional features

Consider to the defect of traditional features which have high empirical components. A vehicle classification algorithm based on the fusion of higher-layer features of a deep network and traditional features was proposed. Firstly, the traditional features of PHOG and LBP-EOH were extracted. Secondly, the higher-layer features excavated from the vehicle pictures by deep belief networks were added, making these three kinds of features together by feature fusion. Finally, support vector machine is used to train and classify the vehicle. When the number of training samples is large enough, the algorithm has a significant effect compared to those with traditional features. It can achieve the accuracy of 95% in the six categories of vehicles.