| With the increasing emphasis on data security and information security,research on network intrusion detection methods can timely prevent and counteract issues such as data leakage,avoiding severe consequences.This article conducts an in-depth investigation into convolutional neural networks and recurrent neural networks in deep learning,as well as the current research status of intrusion detection both domestically and internationally.It is discovered that existing intrusion detection techniques have limitations in classifying high-dimensional and complex network anomaly traffic data.Moreover,common intrusion detection datasets suffer from imbalanced data class distributions.Therefore,this article focuses on two aspects: the research of intrusion detection methods based on deep learning fusion models and data augmentation.Firstly,considering the large cardinality and complex feature composition of common intrusion detection datasets,this article employs convolutional neural networks for feature extraction.However,traditional convolutional neural networks encounter the degradation problem of deep networks,and a single convolutional neural network is insufficient in extracting spatiotemporal features from the dataset comprehensively.To address these issues,this article introduces ResNet residual networks,which use direct mappings to connect different layers of the network,ensuring that information is not lost during the forward propagation process as the number of layers deepens.LSTM networks are employed to learn long sequence time dependencies using three logical control units: input gate,output gate,and forget gate.By resolving the problem of network degradation and utilizing ResNet and LSTM networks for spatial and temporal feature extraction respectively,this article fuses the extracted features and performs classification prediction.Secondly,common intrusion detection datasets currently suffer from imbalanced data class distributions and low data quality.To tackle these problems,this article proposes a data augmentation method based on SGAN semi-supervised generative adversarial networks.This method utilizes a generator to generate pseudo-realistic data based on the real intrusion detection dataset and continuously trains it against a discriminator to generate high-quality data.This process enhances the distribution balance and data quality,while also incorporating attention mechanisms in the model to automatically extract important features and improve their weightage.Finally,this article validates the proposed ResNet-LSTM fusion model and the data augmentation method based on generative adversarial networks using three common network intrusion detection datasets: NSL-KDD,UNSW-NB15,and CIC-IDS2017.Experimental results are analyzed.The results demonstrate that compared to traditional machine learning algorithms and other contemporaneous thesis,the proposed model method has improved accuracy,accuracy,and recall on the NSL-KDD dataset by over4.67%,6.18%,and 3.13%,respectively;Improved accuracy by over 7.2% on the UNSW-NB15 dataset;The accuracy has been improved by over 3.59% on the CIC-ISD2017 dataset. |
