Design And Implementation Of Short Video Classification System

In recent years,the number of Internet videos has grown explosively.Among them,short video has the characteristics of short length,high spread speed,low production cost and strong participation,which make relevant business areas rise rapidly.With the popularization of 5G communication network and the extension of the trend of fragmentary use of the Internet,the short video market will continue to expand in the future.In this context,how to classify the massive short videos quickly that from different sources to effectively manage them is the key point in the short video business.The traditional way of video classification is manual tagging,which includes adding labels actively by users when they upload a video and adding labels manually by business staff after they have watched a video.The former needs the cooperation of users and has low authenticity;the latter takes a lot of manpower and time although it has high accuracy and authenticity.The traditional method is difficult to meet the ever-growing business needs.Therefore,a short video classification system is designed and implemented in this subject in which videos can be automatically classified into multiple categories by algorithms based on their visual content.Furthermore,the system provides a series of management functions such as video management,label management and user information management,which support manual tagging to correct the algorithm classification results.In order to achieve the purpose of automatic classification,this paper proposes and builds a pipeline which executes video preprocessing task and classification task sequentially and automatically.Firstly,video visual features are extracted by the CNN trained in a large-scale image dataset,then these features are encoded by a vector of locally aggregated descriptors,and finally the classification results are output through a neural network.Based on the above pipeline,this paper analyzes the requirements of the system and determines the specific functions that the system needs to implement.According to the result of requirement analysis,the system is divided into following modules:file storage module,upload module,management module,video preprocessing module and classifier module.In the end,this paper tests the core functions of the system and the performance of the algorithm classifier.Based on the analysis of experimental results,it is proved that the system can automatically classify short videos into multiple categories by using algorithms without human intervention,and has higher precision.The short video classification system implemented in this paper can complete the classification task automatically by using algorithms,and also offer manual tagging support so that the system user can correct the classification results of algorithms.The application of this system not only saves a lot of manpower and time,but also ensures the accuracy of classification.