Design And Implementation Of QUIC Traffic Classification Method Based On Multimodal Deep Learning

QUIC encryption protocol has stronger encryption and different flow characteristics compared with other mainstream encryption traffic protocols,which makes the existing single-modal encryption traffic classification methods based on traffic payload or flow characteristics not suitable for the classification of QUIC traffic and has the problem of low accuracy.To solve the above problems,this paper proposes a QUIC flow classification method based on multimodal deep learning.Firstly,the algorithm extracts the statistical characteristics and the payload of the flow as input.The statistical characteristics of QUIC flow are selected by the filtered feature selection algorithm based on feature ranking;Then,the feature views of the two modes are learned by deep learning based on CNN;Then,the fusion layer fuses the feature views,and continues to learn the fused multi-modal features for classification by deep learning,so as to obtain the classification results.The experimental results show that the classification accuracy of the classification model proposed in this paper can reach more than 99%under the self built QUIC flow data set and the public QUIC flow data set,and the accuracy is improved by about 4%compared with the single-mode flow classification method.Based on the above QUIC flow classification method,this paper designs and implements a QUIC flow analysis system.The system mainly includes four modules:data acquisition,data processing,flow classification and data display.The data acquisition layer is responsible for collecting QUIC flow data and shunting it;The data processing layer needs to complete the extraction of statistical features of QUIC flow,the transformation of payload gray image and the analysis of QUIC flow;The flow classification layer is responsible for the classification of QUIC flow application types;Finally,the data display layer displays the QUIC flow stored in the database.This paper tests each module of the system,and the results show that the overall function of the system achieves the expected goal.