Research And Implementation Of Intrusion Detection Algorithm Based On Deep Convolutional Neural Network

With the continuous updating and advancement of network technology,the importance of network security to network users and operators is increasing.Intrusion detection systems play a vital role in maintaining network security.However,due to the application of new network technologies and new devices,traditional intrusion detection algorithms cannot fully extract features in network connections,so the detection rate is low and the error rate is high.Utilizing deep learning powerful data processing capabilities and feature extraction capabilities,it can be applied to intrusion detection to improve the detection capability of intrusion detection algorithms.The research content of this thesis is as follows:1.The KDDCUP99 data set is preprocessed,and an intrusion detection algorithm based on LeNet-5 is studied to transform the intrusion detection into a two-dimensional image recognition problem.The data was input into the above structure for the two-class and five-class experiments,and the final experimental results were obtained.2.The structure of LeNet-5 is improved,and the Inception structure is introduced into the convolutional neural network.A multi-scale CNN-based intrusion detection algorithm is obtained and compared with the previous LeNet-5 algorithm.3.The traditional pooling method of CNN is improved,and a dynamic adaptive sampling method is adopted.The adaptive sampling method is used to dynamically sample different pooled domains,which reduces the loss caused by traditional sampling methods on the sampling of intrusion features.4.Combined with the Inception structure and the residual network structure proposed by Microsoft,Investigate a deep convolutional neural network intrusion detection algorithm to enhance the feature extraction ability by deepening and widening the network structure.Intrusion detection algorithms are evaluated using indicators such as accuracy,false positive rate,false negative rate,and recall rate.The experimental results show that the accuracy of the improved algorithm can reach 94.37%,the false positive rate is only 2.14%,and its performance is improved compared with the traditional machine learning method and DBN algorithm.5.Establish the final deep CNN intrusion detection algorithm and apply it to the actual network for verification.