Research On Intrusion Detection Technology Based On Deep Learning

With the rise of artificial intelligence,big data,cloud computing,5G and other intelligent technology upsurge,various fields are stepping up the deep integration of driving and intelligent technology,but the rapid change of intelligent technology will also bring negative effects,and the network is vulnerable to the invasion of criminals.At present,the current network data volume is increasing,network intrusion behavior is more and more complex,the traditional intrusion detection methods can’t meet the actual needs of the current,deep learning has excellent feature extraction and self-learning ability.In the case of large amount of data and complex data features,it can better play its own advantages.Therefore,this paper applies deep learning to intrusion detection technology,which has better performance than traditional intrusion detection methods.Aiming at the problem of low accuracy of traditional intrusion detection methods,an intrusion detection method based on TBCNN-GRU(two branch convolutional neural network and gated cyclic unit network)is proposed.This method is mainly used for deep intrusion detection of the network,with high accuracy,but the detection time has a slight impact.Firstly,this method creatively transforms one-dimensional network traffic data into three-dimensional gray-scale image data.The spatial characteristics of network traffic data are extracted by two branch convolutional neural network(TBCNN),and the same data sample is coarsened by different convolution layers using the characteristics of its branch structure And refinement extraction,not only retains the overall characteristics of the data,but also continuously mining the deep features of the data;Then,using the advantage of the sequential sensitivity of the gated cyclic unit network(GRU),the temporal characteristics of network traffic data are mined;Finally,the comparative experimental analysis is carried out.The analysis shows that the accuracy of this method reaches 99.88%,and the F1 value is 0.98,which is not only high in accuracy,but also stable in model,and has good effect in improving the accuracy of intrusion detection.Aiming at the problems of traditional intrusion detection methods,such as more parameters and high computing cost,an intrusion detection method based on DSCNN-Bi LSTM(deep separable convolutional neural network and bidirectional long-term and short-term memory network)is proposed.This method is mainly used for rapid intrusion detection of the network,and the detection time is short,but the accuracy is slightly affected.Firstly,the feature dimension of network traffic data is reduced by principal component analysis(PCA),and the one-dimensional network traffic data is transformed into three-dimensional color image data innovatively;Then,the depth separable convolution neural network(DSCNN)is used to extract the spatial features of network traffic data,and the depth separable convolution is introduced to replace the traditional convolution,which greatly reduces the model parameters and makes the network traffic more stable The model is lightweight,and the extracted spatial feature information is transformed into one dimension and then input into Bi LSTM.Taking advantage of Bi LSTM ’s two-way flow of information,the temporal characteristics of network traffic data are mined by taking full account of the influence of before and after features;Finally,comparative experimental analysis is carried out.The analysis shows that the accuracy rate of this method reaches 99.26%,which reduces the model parameters without losing too much precision,and has good effect in improving the efficiency of intrusion detection.In this paper,the traditional intrusion detection methods are improved from the above two directions,and two new intrusion detection methods are proposed to improve the performance of the intrusion detection system.