Research On Intrusion Detection Model Based On Convolutional Neural Network

Nowadays,computer Internet technology has penetrated into every aspect of people's life and become an indispensable part of life.While computer network brings convenience to life,various security threats are also increasing,such as network viruses,trojans,network crimes and malicious attacks by hackers.It is inevitable to rely on computer network to complete electronic transactions in daily life,which to some extent increases the importance of network and information security.Intrusion detection is an active defense method against possible threats in computer networks,which can effectively detect possible attacks and take relevant preventive measures to avoid significant harm to the network.This paper proposes an intrusion detection model based on convolutional neural network,which can protect the networks from security risks by detecting various behaviors.The work of this paper mainly includes the following contents:(1)A batch-normalized convolutional neural network model(BN-CNN)is proposed,which adds batch-normalized processing of data in each layer of convolutional neural network,and then obtains the final classification result through the full connection layer of the network.The batch normalization of data in the model can change the variance size and mean position of data in each layer,which can not only ensure that the data is still original distribution,but also make the data retain the characteristics acquired through neural network learning when input to the next hidden layer.(2)An intrusion detection model based on the focal loss function(FL-CNN)is proposed,which uses the convolutional neural network to train the intrusion detection data and uses the focal function as the loss function of the model.This model can solve the problem of imbalanced data distribution in the training process,reduce the impact of imbalanced data on the model detection results,and improve the accuracy of classification results by reducing the loss of correctly classified samples and focusing more on the wrong samples that are difficult to be classified.(3)A method of intrusion detection data pre-processing is proposed,which converts the original vector format data into n?n image format data,combines the first two models,and then inputs the processed image format data into the intrusion detection model,giving full play to the advantages of convolutional neural network in image processing,so as to improve the performance of the intrusion detection model.The paper carried out the experiment on the NSL-KDD data set and UNSW-NB15 data set and compared the proposed model with the convolutional neural network intrusion detection model without any processing.The experiment verified that the performance performance of the proposed model was better and could improve the accuracy and detection rate of the system while reducing the false alarm rate of the system.