Research And Implementation Of Video Big Data Analysis System Based On Distributed Deep Learning Framework

With the rapid development of the Internet,video data,as one of the main ways for people to obtain information,occupies an increasing proportion in the Internet.There are various forms of video data,such as movies,TV dramas,short videos and so on.Video information contains a wealth of people,events,scenes and other related information,how to use these information for specific content mining to provide traversal for the actual production and life has become one of the hot spots of people's research.In recent years,deep learning has shown significant advantages and potential in target detection and tracking,face recognition,action recognition and other tasks,and is expected to be competent for more complex tasks;at the same time,spark parallel computing framework has achieved remarkable results in classification,clustering and other tasks.Using deep learning model and spark parallel computing framework to mine more accurate and effective related information has gradually become a new focus.At the same time,face exchange video synthesized by artificial intelligence has become a new problem.False video is more and more difficult to identify,which brings a series of challenges to social security.With the continuous progress of video tampering technology and the continuous improvement of video quality,false detection has also brought great challenges.Based on the above background,the main work of this project is as follows(1)This paper proposes an embedded regularization method based on clustering to detect deep fake video.This method takes xception as the basic network,and optimizes the basic network from the network structure,parameter configuration,data set characteristics and the problems studied.The most important optimization part is to increase the simulation of the process of generating fake video,which improves the detection accuracy of depth fake video by detecting the artifacts generated by resolution scaling in the process of generating fake video.(2)The detection algorithm of parallel depth fake video is designed and implemented.With the development of multimedia technology,massive video is emerging,and the amount of data generated every day has reached Pb level.If we process these video data quickly,it will become a research hotspot in the field of big data.In this paper,we design and implement a parallel detection algorithm based on distributed deep learning framework.(3)A video big data analysis platform based on distributed deep learning framework is designed,and the corresponding prototype system is implemented.The analysis platform allows users to use the commonly used video processing algorithms and the depth false video detection algorithm proposed in this paper to solve practical problems.