Research And Implementation Of Semantic Labeling System For National Costume Pattern Based On Large Label Space

The national costume pattern has a long history and rich cultural connotation,which is the great crystallization of Chinese national culture.Integrating technology and culture to explore the culture behind national costume pattern is a topic that has been widely mentioned.This paper takes national costume pattern as the research object,carries out the research on multi-label classification based on the problem of large label space,and develops and implements a set of intelligent collaborative semantic labeling system.The main research contents include:(1)Construction of national costume pattern label system and data set.Through the analysis of the characteristics and content of national costume patterns,a set of label system related to national costume patterns is formed,and on the basis of this system,a set of national costume pattern data set is constructed.(2)In order to solve the problem of large label space,the classical algorithm and label embedding algorithm are experimented on different multi-label classification evaluation indexes.The similarities and differences as well as advantages and disadvantages of different algorithms are analyzed to explore the algorithm direction with certain advantages in the pattern object of national costume.(3)A multi-label classification algorithm based on an improved label embedding method is proposed.First,the deep learning method is used to extract the features of national costume pattern images.Second,the label network is constructed to express the relationship between labels,and the weight representation of the label network is optimized.Thirdly,Laplace dimension reduction is used to compress the label network and preserve the original structure of the label network.Finally,ensemble learning and multi-label classifier are used to automatically label national costume pattern.(4)Combined with the multi-label classification algorithm of costume clothing patterns,an intelligent collaborative crowd-sourcing semantic labeling system with multi-field participation,experts,the public and machines is constructed.The system takes intelligent collaborative crowd-sourcing as the core,combines data retrieval,identity authentication and other functions,and provides users with cultural knowledge output.In order to explore and excavate the cultural connotation of na tional costume,this paper constructs a set of national costume patte rn label system and data set.On this basis,the experimental results of various multi-label classification algorithms in national costume p attern are compared,and then an improved multi-label labeling algo rithm is proposed,and all the results are applied in the semantic m ulti-label learning system of national costume pattern by engineering means.