Research On Network Audio And Video Data Recognition Technology

With the continuous advancements in internet technology and the constant outpouring of web applications,the proportion of data flow in network has changed largely.Among them,due to the high speed of propagation and large volume,network traffic of the videos and audios undermines the quality of network service to a certain degree.Therefore,accurate traffic identification and classification is the key to avoid network congestion and conduct effective network management and resource utilization.How to improve the accuracy of traffic identification and identification efficiency is an important research subject in this field.Therefore,this thesis analyzes and studies relevant technologies from the perspectives of data packet content matching and machine learning method of traffic statistical characteristic.For the existing regular expressions in packet content matching process,as the number of expressions increases,all the expressions need to be matched in sequence,resulting in performance degradation,a regular matching model based on WM(Wu-Manber)improved algorithm is proposed.In this model,the keywords of different protocols are firstly mapped with the corresponding expressions,the keywords are identified by using the WM improvement algorithm,and then only for the content that contains the keyword to match the corresponding expression,in this way,the regular matching can be selectively performed to improve the efficiency of regular matching effectively.Due to the short length of the feature keywords,we analyzed the shortcomings of the WM algorithm when dealing with these keywords and optimized it,the effectiveness of the optimized algorithm is verified through experiments.To solve the problem of network data flow feature selection in machine learning method,First,by analyzing the instability of the weight coefficient of information redundancy among features in MIFS(Mutual Information based Feature Selection)algorithm,and then puts forward the algorithm ARMIFS optimized by means of the linear relationship between features.Through the analysis of the statistical characteristics of different video services,designs a hierarchical k-Nearest Neighbor model that provides fine granularity classification for network video service.In this model,the imported features in each layer correspond to the result of ARMIFS feature selection of all the video services in each layer.The experimental results show that the selected features and the model can effectively improve the accuracy of each video service classification.