Research On Intrusion Detection Technology Based On Densely Connected Convolutional Neural Network

Intrusion detection technology plays a vital role in network security.At present,network attack methods are emerging in an endless stream,and new intrusion detection technology needs to be studied urgently.The application of convolutional neural networks to the field of intrusion detection technology has been widely recognized.It is generally believed that the deeper the network structure,the more accurate it is in terms of feature extraction and detection accuracy.But it is also accompanied by problems such as gradient dispersion,insufficient generalization ability,and large parameter quantity,and low accuracy.In view of the difficulties in the current intrusion detection technology field,this paper introduces an intrusion detection model based on densely connected convolutional neural network,and conducts in-depth research and improvement of the network model.The main work of the thesis includes:(1)In this paper,the densely connected convolutional neural network structure is applied to intrusion detection,and the best densely connected network structure is obtained through experimental exploration.The experimental results show that the KDD99 data set has a depth of 102 and a width of 32 of the densely connected network structure to achieve the best performance.In order to solve the problem of large parameter quantity and gradient disappearance during the training process.(2)This paper improves the data preprocessing method,including the data digitization(one-hot encoding),normalization,matrix conversion,matrix enlargement(bicubic interpolation)to transform the original data set into the most suitable neural The data input from the network can solve the problem of low detection rate.(3)This paper introduces the definition of mixed loss function,proposes to improve the defects of the loss function module in the convolutional neural network,and combines softmax loss with the improved center loss to apply to the proposed intrusion detection of densely connected convolutional neural network in.To achieve the purpose of improving the model index,the recognition rate of new attacks is high.(4)Model compatibility performance experiments were conducted on the NSL-KDD data set different from the KDD99 data set.The test results showed that both of the above data sets have good performance.The paper reveals that the intrusion detection model proposed in it has a good detection effect on Net Flow common abnormal data and widespread attack types,and it also has detection capabilities for new attack data.