A Deep Learning-based Method For Fine-grained Classification Of Network Traffic

Network traffic classification is a part of traffic analysis which plays an important role in improving service quality and managing network resources.With the fast development of network technology and internet services,the network video service grows rapidly.To better distinguish the different qualities of video services and manage network resources,it is necessary for fine-grained traffic classification.It can help network service providers to improve the Qo S(Quality of Service)and the quality of experience(Qo E).In this work,a feature augmentation method is proposed which solves the problem of insufficient input information.First,the bidirectional information and cross information that get from proposed method are transformed into pictures.Then,convolution neural network is used to classify the network traffic represented by pictures.The experimental results show that this method can effectively classify network traffic,and the accuracy rate is 98.58%,which is better than the other two deep learning-based traffic classification methods.The possibility of classifying network video traffic with different resolutions have been explored.By analyzing the collected 480 p,720p and 1080 p data of live and streaming video services,the differences between subcategories are clarified,such as packet density distribution,packet number,etc.On this basis,the proposed method is used to achieve fine-grained classification of video services,and the accuracy got 92.8%.In addition,this thesis also explores the impact of different time scales on the video classification.The results show that when using video traffic data of more than 60 s for classification,the accuracy can reach more than 90%.