Research On Key Technologies For Refined Identification Of Anonymous Network Traffi

Anonymous communication systems,which can hide user identity information,are widely used in illegal activities to evade network censorship,posing great challenges to network security regulation.Anonymous traffic classification is an important aspect of network regulation,but traditional classification methods cannot meet the current traffic classification needs and perform poorly in fine-grained traffic classification.To address these issues,this thesis investigates anonymous traffic at different classification granularities and constructs a hierarchical classification framework for anonymous traffic.Two traffic classification methods are proposed,and the main research content is as follows:(1)In response to the problem of poor performance of traditional machine learning methods in fine-grained classification of mixed anonymous traffic,a hierarchical traffic classification method based on multiscale convolution and efficient attention mechanism is proposed.This method includes coarse-grained classification and fine-grained classification stages.The coarse-grained classification stage quickly filters background traffic using dimensionality-reduced statistical features.The fine-grained classification stage adds multiscale convolution modules to the traditional convolutional neural network to extract more comprehensive and representative features.Meanwhile,an efficient attention module is introduced to capture the relationship between feature channels,thereby learning features with higher relevance and assigning them more weight.Finally,a multilayer perceptron is used to process abstract high-level features,further enhancing the representation ability of traffic categories.Experimental results show that this method can quickly and accurately complete fine-grained traffic classification.(2)Aiming at the problem that the existing single-modal methods do not fully extract anonymous traffic features and the model parameters are difficult to update,an online traffic classification method based on multi-modal self-attention features was proposed.The method consists of three parts: multi-modal input module,feature learning module and online update module.Firstly,three traffic input modes are designed,which can more comprehensively characterize the traffic types.Secondly,different neural networks are used to extract statistical features,packet length sequence features and spatio-temporal fusion features,and the multihead attention mechanism is used to capture the relationship between features in different positions and assign more weights to important features.Finally,the incremental learning mode was adopted to improve the loss function and introduce a deviation correction layer to adapt to the new traffic categories.Experimental results show that the proposed method not only improves the performance of traffic fine classification,but also has the ability to update parameters online to adapt to new traffic categories.