Research On Active Network Flow Watermarking Attack Detection Method Of Tor Network Based On Machine Learning

Tor network is an anonymous communication network realized by onion routing technology.It adopts multi-hop proxy,traffic obfuscation,message encryption and other methods to guarantee the confidentiality and integrity of data transmission and hide the communication relationship between the sender and the receiver.It is one of the anonymous communication networks with abundant users and the most mature technology.However,active network flow watermarking attack can still seriously threaten the anonymity of Tor network.The active network flow watermarking technology changes the behavior mode of the original communication traffic,so that the traffic behavior carries special marks,and uses the watermark detector at the receiving end to detect the traffic behavior containing special marks,so as to judge the communication relationship between the two parties.Compromise the anonymity of the Tor network.Therefore,how to accurately detect stream watermarking in Tor network traffic and improve the anonymity and security of Tor network has become one of the current research hotspots.However,due to the influence of Tor network confusion mechanism,buffer queue,onion message and other factors,the detection accuracy of watermark detection method is low,and the generalization is weak.Therefore,this thesis proposes the detection method of Tor network watermarking by using machine learning and deep learning.The details are as follows:(1)Proposed watermarking detection method of Tor network based on machine learning.In view of the key characteristics of Tor traffic containing network watermarking,this thesis analyzes the time attribute,content attribute and flow direction attribute of Tor traffic from multiple perspectives,and proposes five basic characteristics to represent the traffic and verify its validity.On this basis,23 effective statistical characteristics are expanded.Using multi-type watermark flow as data set,the watermark detection model is trained by machine learning method to realize effective detection of different types of watermarks.The experimental results show that the machine learning algorithm has strong effectiveness and feasibility in Tor network watermark detection,and the detection accuracy is significantly improved compared with the existing watermark detection methods.(2)Proposed the Tor network watermarking detection method based on residual neural network.On the basis of effective statistical features,non-statistical features were introduced to participate in model training to depict watermark flow more accurately.In order to balance the contribution of different features to the model,features are extended by means of replication interpolation.In addition,in order to integrate different types of features,the problem of watermark detection is converted into an image classification issues,statistical features and non-statistical features are combined into a feature matrix,the image conversion module is used to generate gray maps,and the residual neural network model is built to realize watermark recognition.The network can deeply mine a variety of network flow printing image features.The experimental results show that the watermark detection can be realized through watermark image,and the residual neural network model has certain improvement on Tor network watermark detection accuracy compared with the machine learning algorithm.(3)The Tor network watermarking detection system is designed and implemented.Firstly,the overall architecture is designed according to the system objectives,and the functions of each module are described.Secondly,two types of watermarking detection functions are designed according to different user scenarios,and the specific processes of different modes are described.Finally,the functions of each module and the visual graphical interface are realized by programming.