Design And Implementation Of Image Labeling System Based On Machine Learning

With the development of science and technology,humans have entered the era of artificial intelligence,more and more products are towards "intelligence".Artificial intelligence algorithms cannot be separated from the construction of large-scale data sets,so a variety of open source data sets emerge at the right moment.The need for specific data sets provides new requirements for data labeling systems.In addition,data labeling is a tedious and time-consuming work.How to use existing technologies to improve the efficiency of labeling also brings challenges to the design of labeling system.In order to meet the specific requirements,this thesis designs and implements an image labeling system based on machine learning.The system can not only meet the requirements of manual labeling,but also use mature computer vision related algorithms to complete initial automatic labeling and manual modification,thus greatly improving the efficiency of labeling.It's worth mentioning that the system has good expansibility,which lays a good foundation for the preparation of image data set.This thesis firstly introduces the research background and significance of the subject and determines the main work of the labeling system.Secondly,the related technologies and open source libraries used in system design and implementation are introduced and analyzed.Thirdly,this thesis introduces the design and analysis of the algorithms used in image auxiliary labeling,improves the existing human behavior recognition algorithms,and evaluates them.Fourthly,this thesis describes the requirements of the system,introduces the overall architecture design of the system,and divides the system into sub-modules such as image preprocessing,human body node labeling and sports stage classification labeling.Combined with time sequence diagram and flow chart,the general design of each sub-module is described.Fifthly,on this basis,the detailed design and implementation of each module are introduced,and the class diagram and data table are used to show the method implementation in the class.Sixthly,the functions of the labeling system are tested,and the expected and actual results are described.Finally,this thesis summarizes the completed work and makes prospects for the future work of the image labeling system.