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

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.