[1]覃匡宇 唐海洋 谢霄阳.基于改进EfficientNetV2的网络流量分类方法研究[J].大众科技,2023,25(12):1-5.
　Research on Network Traffic Classification Method Based on Improved EfficientNetV2[J].Popular Science & Technology,2023,25(12):1-5.
点击复制

基于改进EfficientNetV2的网络流量分类方法研究()
分享到：

《大众科技》[ISSN:1008-1151/CN:45-1235/N]

卷:
25
期数:
2023年12
页码:
1-5
栏目:
信息技术与通信
出版日期:
2023-12-20

文章信息/Info

Title:
Research on Network Traffic Classification Method Based on Improved EfficientNetV2
作者:
覃匡宇 唐海洋 谢霄阳
（桂林电子科技大学，广西 桂林 541004）
关键词:
网络流量分类; 改进EfficientNetV2; 迁移学习; 超参数优化
Keywords:
network traffic classification;  improved EfficientNetV2;  transfer learning;  hyperparametric optimization
文献标志码:
A
摘要:
针对传统的网络流量分类方法效率低下、对数据集的需求量依赖过大、极度耗费计算资源等问题，文章提出了一种基于改进EfficientNetV2的网络流量分类方法，运用迁移学习的方法，把在大型数据集ImageNet训练达标的预训练模型EfficientNetV2迁移至网络流量数据集进行实验，并依据网络流量数据的特点，将原有网络的输入分辨率进行合理的缩放，在缩短数据训练时长的同时提升了整体精确度；进行多次超参数优化实验后，选用Adam（Adaptive Moment Estim afion）作为优化器并加入CosineAnnealing-Warm-up策略。实验结果表明：改进EfficientNetV2与ResNet50模型、原生EfficientNetV2相比，准确率分别上升了1.19%和1.21%，且模型整体训练时长分别缩减了11 min和5.5 min，在缩短数据训练时长的同时，实现了网络流量的精准分类。
Abstract:
In view of the low efficiency of traditional network traffic classification methods, excessive dependence on data sets and extreme consumption of computing resources, this paper proposes a network traffic classification method based on improved EfficientNetV2. Using the method of transfer learning, the pre-training model EfficientNetV2 which reaches the standard of ImageNet training in large data sets is transferred to network traffic data sets for experiments, and according to the characteristics of network traffic data. The input resolution of the original network is scaled reasonably to shorten the data training time and improve the overall accuracy at the same time. After several hyperparametric optimization experiments, Adam (Adaptive Moment Estim afion) is selected as the optimizer and CosineAnnealing-Warm-up strategy is added. The experimental results show that compared with the ResNet50 model and the native EfficientNetV2, the accuracy of the improved EfficientNetV2 increases by 1.19% and 1.21% respectively, and the overall training time of the model is reduced by 11 min and 5.5 min, respectively. While shortening the data training time, the accurate classification of network traffic is realized.

参考文献/References:

[1] 中国互联网络信息中心（CNNIC）在京发布第 48 次《中国互联网络发展状况统计报告》[EB/OL]. (2021-08-27) [2022-12-27]. http://www.cnnic.cn/gywm/xwzx/rdxw/ 20172017_7084/202109/t20210923_71551. html.[2] Protocol registries[EB/OL]. (2017-08-02)[2022-12-27]. https://www.iana.org/protocols. [3] MOORE A W, PAPAGIANNAKI K. Toward the accurate identification of network applications[C]// International Workshop on Passive and Active Network Measurement, Berlin Heidelberg: Springer, 2005: 41-54.[4] SANVITO D, MORO D, CAPONE A . Towards traffic classification offloading to stateful SDN data planes[C]// 2017 IEEE Conference on Network Softwarization. Bologna, Italy. IEEE, 2017: 1-4.[5] WANG P, YE F, CHEN X J, et al. Datanet: Deep learning based encrypted network traffic classification in SDN home gateway[J]. IEEE Access, 2018, 6: 55380-55391.[6] 熊刚，孟姣，曹自刚，等. 网络流量分类研究进展与展望[J]. 集成技术，2012(1): 32-42.[7] 柏骏，夏靖波，吴吉祥，等. 实时网络流量分类研究综述[J]. 计算机科学，2013(9): 8-15.[8] 刘纪伟，赵杨，李绍晖. 一种基于改进K-means算法的网络流量分类方法[J]. 电子技术应用，2017(11): 86-94.[9] ABBASI M, SHAHRAKI A, TAHERKORDI A. Deep learning for network traffic monitoring and analysis (NTMA): A survey[J]. Computer Communications, 2021, 170: 19-41.[10] WANG W, ZHU M, ZENG X W, et al. Malware traffic classification using convolutional neural network for representation learning[C]// 2017 International Conference on Information Networking (ICOIN). Da Nang, Vietnam. IEEE, 2017: 712-719.[11] WANG W, ZHU M, WANG J L, et al. End-to-end encrypted traffic classification with one-dimensional convolution neural networks[C]// 2017 IEEE International Conference on Intelligence and Security Informatics (ISI). Beijing, China. IEEE, 2017: 43-48.[12] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA. IEEE, 2016: 770-778.[13] TAN M, LE Q V. EfficientNetV2: Smaller models and faster training[EB/OL]. (2021-06-23)[2023-01-15]. https://arxiv.org/ abs/2104.00298.[14] TOUVRON H, VEDALDI A, DOUZE M, et al. Fixing the train-test resolution discrepancy: FixEfficientNet[EB/OL]. (2020-11-18)[2023-01-15]. https://arxiv.org/abs/2003.08237v4.

备注/Memo

备注/Memo:
【收稿日期】2023-03-27【作者简介】覃匡宇（1974－），男，广西马山人，桂林电子科技大学网络与信息技术中心副主任，研究方向为软件定义网络、网络安全。【通信作者】唐海洋（1994－），男，吉林梅河口人，桂林电子科技大学计算机与信息安全学院硕士研究生，研究方向为网络安全。
更新日期/Last Update: 2024-03-04