Research Of Smart Home Intrusion Detection Algorithm Based On Data Augmentation

In recent years,with the development of 5G communication technology and artificial intelligence technology,the Internet of Things has developed rapidly and is widely used in the field of smart homes.However,while it brings convenience to people’s lives,it also faces various network malicious attacks,and there is a security problem of leaking user privacy during communication.Aiming at the low detection rate and high false detection rate of traditional intrusion detection algorithms when dealing with massive and high-dimensional network traffic data,thesis combines the convolutional block attention module and convolutional neural network to propose a smart home intrusion detection network architecture;In view of the problem of low detection rate of minority attacks caused by unbalanced data,a data enhancement algorithm based on log-cosh conditional variational autoencode generative adversarial network is proposed.By balancing the intrusion data set,the overall performance of intrusion detection algorithm can be effectively improved,especially detection rate against minority class attacks.The main work is as follows:(1)Aiming at the problem of poor performance of traditional algorithms in the face of massive and high-dimensional network traffic data,an intrusion detection network architecture based on convolutional neural network is proposed,which includes network traffic data preprocessing and network intrusion detection.The core of the network traffic data preprocessing part is to convert the traffic data from one-dimensional vector data format to two-dimensional matrix data format,so as to improve the data continuity between adjacent features,which is beneficial to improve the performance of the intrusion detection algorithm.This part also includes character-based feature numericalization based on One-hot encoding,numerical feature max-min normalization,and dimensionality reduction of traffic data features based on stacked denoised autoencoder.The network intrusion detection part integrates the convolutional block attention module and the convolutional neural network to build a smart home intrusion detection model.Among them,the convolutional block attention module can effectively extract network traffic features from the channel and spatial dimensions,and improve the intrusion detection performance of the model;By introducing a batch normalization layer,the problem of model performance deterioration caused by changes in data distribution during model training is improved;By adding a dropout layer,the model is prevented from overfitting during the training process.Experiments show that the intrusion detection model proposed in this paper achieves better results in terms of accuracy,precision,detection rate and F1 score than traditional algorithms.(2)Aiming at the low detection rate of intrusion detection algorithms for minority attacks,a data augmentation algorithm based on log-cosh conditional variational autoencoder generative adversarial network is proposed,by introducing the log-cosh reconstruction loss function in the algorithm,the process of generating and reconstructing traffic data is balanced,and the quality of data generation is improved.And by embedding the data enhancement algorithm into the intrusion detection network architecture,the network traffic data set with multiple attacks is balanced,so as to improve the overall performance of the model and the detection performance of minority types of attacks.Experiments show that the detection rate of the intrusion detection model for R2 L and U2 R minority attacks is increased by 22.84% and 0.5%,and the overall detection accuracy increased by 1.15%,the accuracy increased by 2.37%,the detection rate increased by 1.15%,the F1 score increased by 2.51%,and the false detection rate decreased by 0.97%.The intrusion detection algorithm can be deployed at the smart home gateway,which provides theoretical and technical support for the intrusion detection of smart home.