基于多尺度特征融合的异常网络流量检测  

Abnormal Network Traffic Detection Based on Multi-scale Feature Fusion

作　　者：李岚俊 许青 LI Lan-jun;XU Qing(Ma’anshan University,Ma’anshan,Anhui 243000,China)

机构地区：[1]马鞍山学院,安徽马鞍山243000

出　　处：《河北北方学院学报（自然科学版）》2023年第11期7-10,18,共5页Journal of Hebei North University:Natural Science Edition

基　　金：安徽省高校自然科学重点研究项目(2022AH052713)。

摘　　要：提出一种基于多尺度特征融合的流量检测模型,在CNN模型的基础上通过构造多尺度特征聚合块和多尺度池化层从数据流中融合多尺度特征映射,可以有效地从网络流量提取不同尺度的特征,能够增强网络的表征能力,同时通过残差块的使用,可以加速模型的收敛速度,提高检测精度。与不同模型对比,该模型具备更高的精确率和召回率,在检测异常流量方面表现出色。The adaptive feature capture capabilities of convolutional neural networks(CNN)are remarkable as a model for deep learning.The degradation problem becomes increasingly apparent as the model hierarchy deepens.At the same time,due to the neglect of the multi-scale characteristics of network traffic,it leads to the incorrect classification and detection of abnormal traffic.A traffic detection model based on multi-scale feature fusion was proposed.Based on the CNN model,multi-scale feature maps can be fused from data streams by constructing a multi-scale feature aggregation block(MFM)and a multi-scale pooling layer(MPL).This effectively extracted features of different scales from network traffic and enhanced the network's representation ability.At the same time,the use of residual blocks accelerated the convergence speed of the model and improve detection accuracy to 99.70%.Compared with different models,this model performs well in detecting abnormal traffic.

关 键 词：CNN 多尺度特征 MFM MPL 残差块 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]