Research On Network Encrypted Traffic Identification Technology Based On Deep Learning

The rapid development of Internet has changed people's lives,but the complex network environment also brings a variety of problems.And network traffic identification plays an important role in network service governance.By identifying traffic to allocate network resources reasonably,service quality and user experience can be improved.At the same time,the accurate identification of malicious traffic and abnormal traffic can effectively protect the network system and is very important in maintaining network security.At present,the existing network traffic identification technology includes port based identification technology,deep packet detection based on payload identification technology,payload randomness based identification technology,host behavior based identification technology,machine learning method based identification technology and deep learning method based identification technology.Because of the problems of dynamic port,the popularity of encrypted traffic and the difficulty of feature selection,the former methods can not meet the current needs well.Deep learning technology is used to identify and classify the encrypted traffic on the open dataset,and Keras framework is used to build the model with Tensorflow as the background.The recognition granularity of the model is session level.After the traffic is segmented in session form,the session is transformed into gray-scale image through different processing forms.Firstly,the influence of different traffic processing forms on classification effect and the role of load on classification are discussed under One-dimensional Convolutional Neural Network.Then the Residual Network is used to classify the traffic service types.On the basis of service type classification of traffic,the traffic of the same service type is classified by application type.After classifying all the traffic according to the application types,the application types of 16 types of application traffic are classified.Aiming at the problem of the imbalance of sample data among the categories,the use of the Generative Adversarial Networks to supplement the data of the low sample size categories can improve the classification performance.Experimental results show that this method can effectively identify and classify encrypted traffic.