Network Anomaly Detection For Video Private Network

The application of public security video network（hereinafter referred to as video private network）makes public security work more efficient,and the security of video private network has become a hot topic of research concern.Network anomaly detection technology can effectively detect network anomalies such as camera vulnerabilities and weak passwords,providing support for securing video private networks.Currently,there is a lack of specialized anomalous behaviour data for video private networks,and the common network anomaly detection model for public networks needs to be optimized to meet the specificity of the video private network environment.Accordingly,the main results of this research,in terms of expanding the data and optimizing the network anomaly detection model to improve accuracy,are as follows:Ⅰ.A simulated network environment for public security video private networks is constructed,verification algorithm metrics are proposed and a video private network anomaly detection dataset is established.In order to explore the specific characteristics of network anomalies in a dedicated video network,a network environment was built,penetration tests were conducted and data were obtained.In order to ensure a more realistic sample of the dataset,the original traffic data was collected in a local public security video network and the data was desensitized and cleaned.The hexadecimal information in the data packets was converted into time series to retain more hidden features,while refined features were extracted from the data of the attack target IP address,time,attack method,and other attributes,and the NAD-RVSN-2021 dataset was produced in CSV format.Ⅱ.Two network anomaly detection models are designed and optimized for video private networks.One is a network anomaly detection model that combines TCN with an attention mechanism;the wider perceptual field of the TCN algorithm enhances the model’s ability to mine the spatio-temporal features of video stream data,and the residual blocks in it enable the network to transfer information in a cross-layer manner,enabling deep network training.At the same time,the attention mechanism is used to enhance the model’s attention to the anomalous features associated with different network attacks,further enhancing the model’s effectiveness in discriminating network anomalies.On the NAD-RVSN-2021 dataset,Accuracy was 95.83%improved by 10% compared to baseline models such as CNN;on the UNSW-NB15 dataset,Weighted Precision was 72.39% improved by 8%.Other key metrics also improved significantly over the baseline model on the KDD and UNSW-NB15 datasets.The second is a network anomaly detection model that combines Res MLP,which reduces the number of operations and extracts the spatial features of the data more efficiently,and LSTM,which has a flexible gating unit to improve the model’s ability to extract the temporal features of the video stream data.Accuracy is 95.83% on the NAD-RVSN-2021 dataset,a 10% improvement compared to the baseline model,and weighted Precision is 97.06%,a 3% improvement.On the KDD and UNSW-NB15 datasets,the key metrics performed better than the baseline model.Ⅲ.Design and implementation of a prototype network anomaly detection system for video private networks.The analysis of network anomaly detection requirements specific to front-end devices and camera management systems in video private networks,the integration of the above optimized and designed network anomaly detection model for video private networks,and the visualization of detection results.Validation of the simulated network environment penetration test data and the NAD-RVSN-2021 dataset shows that the system achieves 95%detection accuracy,which is in line with the expected results.