摘要
为研究大型深层卷积神经网络在多类别暗网流量检测领域的适用性,基于Resnet、Densenet与Xception网络进行分类性能比较。将3种模型在Darknet2020暗网数据集上进行验证,使用9.3万条Non-Tor数据、1300条Tor数据、2.3万条Non-VPN数据及2.2万条VPN数据进行实验。结果表明,3种模型均能快速处理海量数据,且对Tor与Non-Tor流量的检测结果较好,F1值最高可达到0.91,但对VPN与Non-VPN的分类效果有待提高。选择在测试集上检测性能最好的Densenet网络,加入GRU网络提取时序特征进行改进后,总体分类精确率为83.4%,召回率为82.2%,检测性能得到进一步提高。
In order to study the applicability of large deep convolutional neural networks in the field of multi-class darknet traffic detection,the classification performance were compared based on Resnet,Densenet and Xception networks,respectively.Put the three models on the Darknet2020 data set for verification,using 93000 pieces of Non-Tor data,1300 pieces of Tor data,23000 pieces of Non-VPN data and 22000 pieces of VPN data for experiments.The results show that the three models can quickly process massive data,and the detection results of Tor and Non-Tor traffic are good,with the F1 value up to 0.91,while the detection results of VPN and Non-VPN need to be improved.Among them,the Densenet network has the best detection performance on the test set.After adding the GRU network to extract the timing features for improvement,the overall classification accuracy rate is 83.4%,the recall rate is 82.2%,and the detection performance is further improved.
作者
崔见泉
周延森
刘博宇
郝嘉琪
CUI Jian-quan;ZHOU Yan-sen;LIU Bo-yu;HAO Jia-qi(Graduate Department,University of International Relations;School of Cyber Science and Engineering,University of International Relations,Beijing 100091,China)
出处
《软件导刊》
2022年第3期176-180,共5页
Software Guide
基金
国际关系学院大学生学术支持计划项目(3262020SYJ007)。
