Detection Algorithm Based On Hybrid Attention Mechanism And Generative Adversarial Network

With the booming development of information technology,the Internet facilitates all aspects of life.However,behind these numerous advantages,there are also many security problems.Although there are many preventive measures to solve part of the problem,such as firewall,information encryption and other technologies,but their processing capacity is limited.As an important security mechanism,network intrusion detection itself can detect suspicious transmission,and take active response measures,greatly improve the security of the network,effectively make up for the defects of the firewall,make the information security infrastructure more complete.At present,many methods have been adopted to complete the intrusion detection task,such as machine learning.Although some effects have been achieved,there are also some defects.For example,most methods only focus on the detection rate of the whole,ignoring the detection rate of each category.At present,there are more studies on dichotomies than on multi-dichotomies.Most scholars ignore the problem of imbalance in the network,which still needs to be dealt with further.In order to solve these problems,this paper proposes an intrusion detection method based on hybrid domain attention mechanism and generative adversal network.Experiments show that this method not only improves the overall accuracy,but also improves the detection rate of a few types of attack samples.The main work contents are as follows:Firstly,in order to solve the problem of poor detection effect in binary classification and improve the overall accuracy of multiple classification,the cascaded network model composed of deep separable convolution and short and long memory network is applied to intrusion detection.The unique advantages of each network can be exploited through series connection,and then the advantages of the two networks can be combined through parallel connection to extract spatial and temporal features of data and improve the detection ability of the model.Secondly,in order to improve the detection rate of each category in the dataset,a hybrid domain attention mechanism is proposed to further update the dimension of features from two perspectives,calculate the importance of features,and generate more accurate feature vectors.After that,it is fused into the cascade model of this paper to make a deep supplement to the extracted features and improve the detection ability.Finally,a data balancing method based on generative adversarial network is proposed to alleviate the low detection rate of a few attack types caused by data imbalance in the data set.The model uses variational autoencoder to replace generator,and the residual network is integrated into generator and discriminator network.A new loss function is introduced and the optimal transmission distance is used as the parameter updating method.This model can effectively expand the original data set,reduce the negative impact of data imbalance,and improve the final detection effect.The model was evaluated based on the CIC-IDS2017,NSL-KDD and UNSW-NB15 data sets.Experimental results show that the accuracy of the proposed method for the above data sets can reach 99.80%,99.32% and 83.87%,which is higher than other commonly used methods.