Research And Implementation Of Website Fingerprint Attack And Defense Technology Based On Deep Learning

With the rapid development of Internet technology,people enj oy the convenience of network services,but also suffer from various network security issues represented by privacy leaks.Anonymous network communication technology can effectively protect the privacy of network users by concealing the identity and communication relationship between the two parties.However,there are also malicious users who abuse the anonymous communication network to post bad information or engage in illegal and criminal activities,causing serious social harm.In order to combat cybercrime and ensure the effectiveness of network supervision,website fingerprint attack technology based on various classification algorithms came into being.This thesis analyses the shortcomings of previous fingerprint attacks and defense technologies,and proposes a fingerprint attack technology based on deep learning website and a website fingerprint defense technology based on adversarial example.A website fingerprint attack technology based on deep learning is proposed.In the past,the website fingerprint attack module used traditional machine learning algorithms to classify.The accuracy rate depends on the quality of feature selection.Model classification accuracy has an upper limit and the defense against the website fingerprint is poor.In this thesis,a deep learning attack model based on ResNet and GRU is proposed.Through its powerful feature learning and relationship fitting ability,it achieves more than 99%classification accuracy on the current largest website fingerprint dataset and can well resist the current mainstream fingerprint defense technology.A website fingerprint defense technology based on adversarial example is proposed.Previous fingerprint defense technology not only caused a significant increase in network bandwidth load,but also could not resist the deep learning websites fingerprint attack.According to the weakness of adversarial example attacks in deep learning model,this thesis proposes a web fingerprint defense method based on adversarial example.The method uses differential evolution algorithm to generate web fingerprints adversarial example perturbation,fills and modifies the web fingerprints,which reduces the classification accuracy of the deep learning web fingerprint attack model by more than 61%,and makes the classification effect of the previous attack techniques also greaty reduced.Finally,based on deep learning website fingerprint attack model and practical application scenarios,this thesis designs and implements an efficient,real-time,interactive and friendly website fingerprint attack system.