Abstract: The rapid development of mobile Internet brings many special problems in the fields of network security, network measurement and quality of service. In order to further study the characteristics of mobile Internet, researchers need to quickly and accurately classify the mobile traffic flow from the traditional network traffic. In this paper, combining lightweight flow table and deep packet inspection(DPI)technology, a real-time mobile network traffic classification approach was proposed. To reduce the scale of flow table, DPI overhead and improves the accuracy of mobile traffic classification, the network flow was expanded into the sequence flow segments according to the interval-time relationship, and the mobile traffic was classified accurately according to DPI of first N packets in the sequence flow segments. The real-time network traffic experiments show that, the identification accuracy rate can reach 91.55%, the average overhead of one DPI only takes 20 packets,and the scale of flow table can be reduced to 0.21%. Compared with the P0F, the accuracy of the propose approach can be improved significantly.