Detection System Of P2P Video Stream Based On Content Recognition

In the current Internet user traffic classification,the multimedia video traffic which occupies the router port traffic is increasing rapidly,accounting for about 70% of the total traffic.In addition to the legitimate multimedia services provided by ISPs under effective supervision,there are also many unregistered P2 P video services.How to effectively detect the content of these explicit and implicit P2 P streams is the main content of this study.Firstly,this paper introduces the background and research purpose of the subject,investigates the main concepts of current P2 P flow technology,P2 P hierarchical structure,main P2 P flow detection technology,and finally proposes a n engineering realizable solution based on Application Protocol feature string.Then,from the perspective of content recognition,this paper investigates various classification methods for known video sets,including support vector machine,softmax regression method,neural network method,and proposes a quasi three-dimensional convolution neural network content recognition method with residual.Combined with P2 P flow detection technology and content recognition method,this paper proposes a P2 P video stream detection(VSD)system,which consists of two parts: front and back.The front-end P2 P flow capture system includes data collection end,data analysis end,P2 P packet analysis module,database form design,client statistical display module,and gives an example of real-time port P2 P flow detection after project implementation.For the P2 P multimedia video samples which have been sampled by the front-end system,the back-end gives the specific implementation method of video stream content recognition,inc luding the input sample video frame pre-processing,typical frame group extraction method(TFS),Parallel spatial and temporal feature extraction,and finally through the improved quasi three-dimensional convolution residual network(Q3D Res Net)for training and comparison to output the sample's classification label.The simulation results show that the improved Q3 D Res Net greatly reduces the operation time,and the accuracy of the common ten classification tags is more than 79%.Considering the social prop erties of multimedia video,this paper also gives a fast content recognition method based on positive and negative energy,which can effectively detect negative energy video samples and provide them to the network behavior supervision for decision.The main innovations in this paper include:(1)in the traditional P2 P detection methods,a P2 P flow detection engineering realizable method based on application layer feature string Application and flow statistics information is proposed;(2)an improved method o f video content recognition based on quasi three dimensional convolution residual network is proposed to effectively improve the accuracy of traditional classification mode and negative energy video fast detection.