Research On Intrusion Detection Based On Generative Adversarial Network And Dual Attention Mechanism

With the continuous development of computer and Internet in recent years,the Internet has brought us great convenience,but at the same time,it is accompanied by the endless emergence of various network security problems,which has caused very serious consequences for society and individuals.The traditional network security defense technology led by the firewall is difficult to face the current severe network security situation,and the intrusion detection model with strong feature learning ability can still show outstanding recognition ability in the face of today’s various network attacks,but the current intrusion detection model still faces the problem of data imbalance caused by the scarcity of training data set samples.This has a great impact on the intrusion detection model in the recognition rate of network attacks,which needs to be improved urgently.Aiming at the problem of low detection rate of minority attack samples when dealing with unbalanced intrusion detection samples in the field of intrusion detection,this paper firstly proposes an oversampling model based on improved conditional generative adversarial network.Wasserstein distance and GELU activation function are introduced into the conditional generative adversarial network.Thus,the problems of training instability and mode collapse in the original generative adversarial network are improved,and the network performance is improved.The improved conditional generative adversarial network is used to oversampling the scarce samples,so as to alleviate the problem of data imbalance.Experimental results show that the model can generate high quality intrusion detection data.The previous intrusion detection models usually use a single way to solve the problem of data imbalance in the intrusion detection data set.This paper proposes to improve the classification model of Dense Net by using dual attention mechanism and GHM loss function to solve this problem.The GHM loss function can make all kinds of samples in the data set contribute to the training of the model more evenly.Dense Net can deepen the network while solving the problem of gradient disappearance,and its special neural network structure also helps to transmit features.Experiments on the NSL-KDD dataset and UNSW-NB15 dataset show that the overall accuracy of the model proposed in this paper reaches 81.1% and 78.3%,respectively,and is superior to other intrusion detection models in terms of both the F1 value of minority attack samples and the overall accuracy.Finally,combined with the actual campus network topology architecture,a campus network intrusion detection system deployment scheme is proposed to improve the current campus network security capabilities,showing the actual application value of this model.The thesis has 32 figures,7 tables,and 63 references.