Research And Implementation Of Pornographic Detection In Video

With the rapid development of network technology,people can easily get all kinds of information and communicate with each other through the Internet.But the information is mixed on the Internet,people may be disturbed by various bad information which may cause people financial loss,seriously may affect people’s mind.In such an circumstance,supervising the content of the network is necessary.But the traditional supervision strategies have obvious deficiencies at the present stage when the scale of Internet users is huge and the network information grows explosively.In order to improve this situation,and provide more efficient,intelligent network content supervision strategy,this paper go in deep study about this problem.Based on the strategy of artificial neural network and late fusion of video frame,this paper introduces a pornographic information recognition in live video model,and verifies the feasibility of this model to identify whether the live video contains pornographic information.In order to improve the performance of this pornographic information detection model in video further,the conclusions in Computational Complexity Theory are applied.Through experiments,the effectiveness of the optimization is verified by comparing the classification accuracy of the model before optimization and the model after optimization on the test dataset.The main content of this paper includes the following aspects:1.Investigate mainstream content supervision technologies in Internet,analyze the pain points of traditional content supervision methods,research technical solutions that can be used to solve the pain points of traditional content supervision,and propose a scheme that apply technology based on deep learning to video content supervision.2.In-depth research on the theories of artificial neural network technology,get familiar with the construction method and training process of artificial neural network from network structure and training strategy,and focus on the application of artificial neural network in video content recognition.3.Explore the use of open source machine learning library Py Torch,and complete the construction and training of artificial neural network with this machine learning library.Analyze the training strategy of the network model based on the characteristics of the experiment,and propose a training scheme based on transfer learning.The pretraining dataset is built based on the open source image dataset.The video data needed for the experiment is collected from the well-known live platform,and sample the video frames to construct the formal dataset that can be trained by the network model.4.Deeply study computational complexity theory and analyze the accuracy of the model after training.On the basis of the video pornographic information detection model designed by this paper is effective in live scenes,the conclusions of the computational complexity theory are applied to the network mode.The results show that the classification accuracy of model after optimization is improved compared with model before optimization.