Deep Fusion Feature for Vehicle Classification and Recognition

To solve the problem of higher energy consumption and lower recognition rate caused by artificial extraction feature, which is sophisticated and single generally in the process of vehicle classification and identification, in this paper, we propose a method for video vehicle processing using deep fusion feature based on deep learning. This approach mixes all of the tail characteristics of vehicles after each Sampling layer together, including global characteristics of low-dimensional and local sparse response of high-dimensional, this allows us to get more information about the features, then softmax classifier is used for classification steps. Experimental results show that the final classification accuracy can be achieved about 99.32% by combining the characteristics of a single treatment with features of various strata after Sampling layers, which called full feature fusion. The algorithm we suggested is applicable to identify highway video vehicles, the recognition speed and accuracy of this method are improved significantly compared with the general means.