基于改进残差网络的异常网络流量检测  

Network anomaly traffic detection based on improved residual network

作　　者：李岚俊 王英明 胡昊 李洁 LI Lanjun;WANG Yingming;HU Hao;LI Jie(School of Big Data and Artificial Intelligence,Ma’anshan University,Ma’anshan 243000,China)

机构地区：[1]马鞍山学院大数据与人工智能学院,安徽马鞍山243000

出　　处：《长春工业大学学报》2023年第5期468-473,共6页Journal of Changchun University of Technology

基　　金：安徽省高校自然科学研究项目重点项目(2022AH052713)。

摘　　要：提出一种改进ResNet101的异常流量数据检测和识别方法,在残差模块中改变卷积层,同时引入高效信道注意力(ECA)机制,使用一维卷积在高度注意力信道中融合特征流,增加对异常流量的识别能力。实验结果表明,基于残差网络改进模型在识别极低样本数量时相比原有模型能够有更高的精确率、召回率和F 1值。In response to the above problems,this paper proposes an improved method for detecting and identifying abnormal traffic data of ResNet101.The convolution layer is changed in the residual module,and the Efficient Channel Attention(ECA)mechanism is introduced.The feature flow is fused in the high attention channel using one-dimensional convolution to increase the recognition ability of abnormal traffic.The improved residual network can identify abnormal traffic with very low sample number.According to the experimental results,the improved model based on residual network can have higher accuracy,recall rate and F 1 value than the original model.

关 键 词：流量数据检测 ResNet101 ECA 不平衡 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]