Research On Intrusion Detection Method Based On PGoogLeNet-IDS Model

With the rapid development of 5G,the Internet,cloud computing,and other technologies,the network environment has become increasingly complex,and the forms of attacks have become increasingly diversified,which has also brought severe challenges to the security of cyberspace.Network intrusion detection technology adopts an active defense method to maintain network security,which can provide real-time monitoring and dynamic protection for the network.In the face of massive network data,the existing intrusion detection models can neither effectively process these high-dimensional and complex data,nor identify the types of attacks with fewer data samples,and the model detection efficiency is poor and the recognition rate is low.Because of these problems,this paper proposes an intrusion detection system based on the PGoogLeNet-IDS model.The main work of this paper are as follows:1.Aiming at the problem that the intrusion detection model does not recognize the attack type and the detection efficiency is low,we propose an intrusion detection system based on the GoogLeNet-IDS model.First,we improve the GoogLeNet model to make it a model that can process network data.Secondly,this article builds a lightweight SE_DSC module based on SENet to replace the Inception module in the original model.This module is a sub-module SENet that introduces the attention mechanism in the depth separable convolution module,where the depth separable convolution completely decouples the channel and the space,and reduces the model's performance without affecting the model detection performance.Training parameters improve the detection speed of the model;and the SENet module improves the feature extraction ability of the model by increasing the focus on the current task-related features and discarding the attention of irrelevant or less useful or useless features.Finally,the performance of the model is verified on the NSL-KDD data set.The experiment results show that the model proposed in this paper has achieved more significant effects in terms of detection speed and detection accuracy.2.Based on the GoogLeNet-IDS model,we continued to analyze the network traffic data samples and found that the distribution of data samples of normal traffic and abnormal traffic is extremely unbalanced.The model is difficult to fully learn the characteristics of the minority samples,resulting in these minority samples The accuracy is low and the model classification performance is poor.For this reason,a PGoogLeNet-IDS model based on unbalanced processing of data distribution is proposed.This model uses the multi-class focus loss function proposed in this paper in the GoogLeNet-IDS model,and improves the model's recognition rate of each attack type by dynamically increasing the weight of minority samples and reducing the weight of majority samples.By viewing the test results on the datasets NSL-KDD and UNSW-NB15,the PGoogLeNet-IDS model proposed in this paper effectively improves the attack on each type,and also enhances the overall classification performance of the model to a certain extent.3.Based on the above-mentioned intrusion detection algorithm,this paper constructs an intrusion detection system based on the PGoogLeNet-IDS model.The system includes important modules such as data management,data preprocessing,algorithm operation,and references to facilitate scholars'specific operation procedures for intrusion detection technology.Common data sets and algorithms,and the model proposed in this article for learning.The system provides user-friendly operation steps and realizes the convenient service of intrusion detection system.