Research On Intrusion Detection Model And Method Based On Channel Attention Residual Network

With the development of network technology,we have entered the information age,and network security has become a key concern for people and the country.Intrusion detection technology is a key technology for maintaining network security.It protects network security by monitoring traffic in real-time and responding promptly to threat events.How to accurately identify malicious attack traffic is a key issue for intrusion detection systems,which scholars have been studying.With the continuous innovation and development of network technology,the network environment has become increasingly complex,and traditional intrusion detection technologies are facing severe challenges.At present,machine learning has shown excellent capabilities in solving classification problems in the fields of image recognition and text classification,and has been used by researchers to construct intrusion detection models.Deep learning can learn more about data features and handle high-dimensional data through deep-seated networks.Therefore,this paper applies deep learning to the construction of intrusion detection model to meet the network environment in which various attack means emerge in endlessly.The main research contents are as follows:(1)Aiming at the problem of low accuracy of traditional intrusion detection model,this paper uses deep learning network to design intrusion detection model.Convolutional neural network is used to learn network traffic characteristics,but with the increase of depth,network degradation will occur.Residual network solves this problem by adding residual blocks.Based on this,this paper uses res2 net to build intrusion detection model.Res2 net uses multi-scale convolution kernel to replace the convolution kernel of RESNET residual structure,which increases the receptive field of the model and can learn network traffic characteristics more finely.At the same time,channel attention mechanism ecanet is added to the construction of the model.Channel attention can make the model focus on important channels and improve the efficiency of the model.(2)In view of the imbalance of samples in each category of the dataset,some categories with small sample numbers cannot be fully learned by the model,resulting in low detection accuracy.In this paper,SMOTEENN comprehensive sampling algorithm is used to balance the data set,and SMOTE algorithm is used to expand R2 L and other categories of data in KDD CUP99 data set,and then ENN algorithm is used to clean samples at the boundary to increase the accuracy of model detection.(3)As the training of the model using high-dimensional feature data sets takes too long,and some features do not play a positive role in the training of the model,this paper uses the feature recursive elimination algorithm(RFE)to select features and select reasonable feature subsets to reduce the training time of the model and improve the accuracy of the model.This paper uses the KDD CUP99 dataset to train and test the model,which can achieve an accuracy rate of 93.8%.By comparing the recall accuracy rate and F-1 value with other models,it proves the feasibility of this model.