Researchon Protocol Recognition TechnologyBased On Deep Learning

With the rapid development of the Internet,communication transmission capabilities havebeengreatlyimproved and different kinds of network applications come out one after another,therefore,our daily life is becoming more and more inseparable from the network.But as a consequence of the expanding network,malicious attack activities such as Trojan program and worm virus have also been upgraded and get expanded,which has seriously affected the experience of the majority of Internet users as well as caused huge losses to social production and life.So,the security of communication in network is now the central issue.In the field of network security,protocol identification is a prerequisite foundation for other technical work such as vulnerability mining and intrusion detection.It is of great significance for ensuring communication security in the network and has become the focus of researchers in recent years.There are a large number of private protocols in nowadays network.These protocols'specifications are not open online,and the traditional protocol identification technologies have some certain defects.The port-based protocol identification technology needs to know the predefined port number,and the current dynamic port technology makes the port not fixed;the load-based and behavior-based protocol identification technology requires a lot of manual operation,and the feature design are complicated.These factors have led to the traditional protocol identification technologyno longermeeting the current high-speed and convenient requirements.In this paper,for the needs of large-scale application layer automatic and accurate protocol identification,the research is carried out based on the methods in deep learning.With reference to the maturetechnology,the convolutional neural network used in the classification and the autoencoder used in clustering,we accomplishedopen protocol classification and private protocol clustering respectively.The detailed research work of this paper is as follows:1.A protocol identification method based on convolutional neural network is studied.The targeted data of this method is theopen protocols,which means protocol classification under supervised learning.The protocol data unitsare converted into a bitmap format which is suitable as the input of the two-dimensional convolutional neural network,and deep-level features are extracted through the convolutional layer and the pooling layer alternately.The experimental results demonstratethat for the six openapplication layer protocols collected in real network environments,after the protocol classification model is trained,the classification accuracy of the test data set can reach 97.30%.2.A protocol identificationmethod based on autoencoder is studied.The targeted data ofthis method is theprivate protocols,which means protocol clustering under unsupervised learning.The loss function in the autoencoder is improved by joining the autoencoder reconstruction error and the maximum likelihood function in GMM clustering together,so that the original data samplesget dimensionally reduced and are projected to the GMM clustering-friendly space.The experimental results demonstratethat for the six privateapplication layer protocols collected in real network environments,the clustering accuracy of the protocol identification model can reach 91.71%.