Research On Modeling And Defense Technology Of Website De-anonymization Attack Based On Traffic Fingerprint

Anonymous network communication method is a network communication method with a high anonymity degree so that it is widely welcomed by users who need anonymity.However,the attacker is still able to mine the leaked private information from the communication traffic data in the anonymous network,and build a corresponding classifier to correlate behaviors of users.The passive attack method of constructing an anonymous network traffic fingerprint model for traffic analysis is a major de-anonymization attack method.Aiming at the problem of insufficient feature extraction in processing long fingerprint sequences by existing deep learning attack models,a TCN-BL classification attack model is constructed.TCN-BL combines time series convolutional network models and bidirectional long and short-term memory networks to further improve the receptive field of the convolutional layer to deal with Longer fingerprint sequence features.It solves the problem of insufficient feature extraction ability in the data sequence at long distances.The model does not verify the model from the closed world scene data and the open world scene data.Compared with the traditional convolutional neural network model,it can show better results in both scenarios,reaching 97.65% and 95.23% respectively.At the same time,the classification effect is analyzed for the training model of data sets of different sequence lengths and sample numbers.To a certain extent,the results show that the TCN-BL model can adapt to the training task of small sample datasets.Through the construction of the deep learning classifier attack model,from an opposing perspective,consider the implementation of defense strategies for the protection of network anonymity.Combining with the deep learning adversarial attack algorithm,it generates specific perturbations for filling to form adversarial samples.For the generated samples,comprehensive evaluation is made from the load consumption caused by the disturbance and the classification accuracy in the attack classifier model.At the same time,this paper also uses the We FDE information leakage analysis system to analyze the information leakage of adversarial samples.The result shows that in the overhead consumption and accuracy analysis,the adversarial samples generated in this article have better results than traditional defensive datasets.The overhead consumption is reduced by 20%,and the accuracy of the attack classifier is reduced by more than 30%.At the same time,the information leakage rate dropped by 0.72 bits on average compared to the more defenseless dataset.