Research On Audio And Video Data Recognition Technology In High Speed Environment

The popularity of Internet technology and the increasing diversification of streaming media applications have led to an increase in the proportion of audio and video services in network traffic,which brings a curshing burden and security risks to the network.How to effectively allocate network bandwidth resources and ensure network security is a key point for Internet Service Procider(ISP)and network supervision departments.As one of the basic technologies to enhance the controllability of the network,the audio and video data identification technology can not only help the ISP to provide different quality of service according to different audio and video services,thereby optimizing resource allocation,and can timely understand the network status and ensure network security.Therefore,audio and video data recognition technology has increasingly become the focus of attention in recent years.This paper mainly studies two mainstream audio and video recognition technologies,namely audio and video data recognition technology based on deep packet detection and audio and video data recognition technology based on machine learning.Aiming at the shortcomings of the depth packet detection method in recognition speed,the paper proposes an improved DPI audio and video recognition model based on the feature library matching algorithm.The model makes the identification more accurate and efficient through session and dynamic session association methods.At the same time,the feature library matching algorithm selects the DFA engine to improve the recognition speed,and selects the grouping algorithm to reduce the state expansion caused by DFA compilation.In addition,in order to improve the matching speed of signature database,the paper studies the impact of traffic environment information on DPI detection performance based on the application layer protocol distribution characteristics and actual network traffic behavior characteristics,and proposes a regular expression grouping algorithm based on protocol priority and adaptive network traffic scheduling.Through comparison experiments,it is verified that the grouping algorithm adopted by the model has a good effect in reducing memory expansion and has an advantage in matching speed over the algorithm of non-priority grouping and non-adaptive network scheduling,and also validates recognition ability of the model.In view of the lack of recognition accuracy of machine learning methods,the paper proposes a hierarchical KNN classification model.Since the performance of the classification is related to the algorithm itself and decreases as the classification category increases,the number of classification categories can be reduced by the layering method,and each KNN classifier can identify a specific audio and video service,thereby achieving a more accurate classification purpose.In addition,finding a subset of features that can truly reflect the essential differences between audio and video traffic is the key to effective classifier classification.Therefore,the paper proposes a feature selection algorithm based on ReliefF and consistency measurement,and selects a combination of features with distinct distinguishing effects for each classifier.The ReliefF algorithm is used to remove some non-critical features to reduce the dimension of the feature,and then the consistency measurement algorithm is used to filter out the redundant attributes.Through this method,a feature subset with discrimination can be obtained to improve the classification accuracy of the algorithm.Through comparison experiments,the advantages of the feature selection algorithm and the hierarchical KNN classification model proposed in this paper are verified.