Font Size: a A A

Research On Intelligent Design Method Of Children's Clothing Based On Machine Learning

Posted on:2021-03-13Degree:MasterType:Thesis
Country:ChinaCandidate:D F QiuFull Text:PDF
GTID:2381330611961975Subject:Engineering
Abstract/Summary:PDF Full Text Request
The garment industry is a representative in large production line,large division of labor,and large-scale production of traditional industries,yet each garment industry can only produce its own style of clothing,which leads to the design technology in clothing limited to the garment brand of the company.With the flow of fashion designers,the company's clothing style design technology will be lost accordingly,so the intelligent design of the garment industry has been greatly valued along with the development of machine learning.This thesis takes science and technology projects as the starting point,establishes children's net body size and children's clothing model size database,uses machine learning technology to establish a neural network model system for children's clothing intelligent design for automatically forming children's clothing size models.And then children's clothing styles will be transferred by reconstructing a super-resolution for low-resolution of children's clothing pictures to obtain high-resolution images of children's clothing,thus improving the effect of children's clothing style transfer.The main work is as follows:First,design an intelligent design system for children's garments size and collect children's net body size data and clothing model size data according to the needs of the actual project;store the collected children's net body size data and clothing model size data in the database;conduct the analyzing and learning training for children's net body size data and clothing model size data in the database above to build a prediction model;enter the net data of the children to be customized,and obtain the detailed specification data for the pattern design of children's clothing according to the prediction model;enter the detailed specification data into a paper pattern model made with CAD to obtain a customized children's clothing pattern,breaking the current situation that traditional children's clothing is mainly patterned according to a fixed height standard body shape,to meet the needs of children of different heightsand body sizes for tailored clothing,and realize the mass design of individualized customized children's clothing fast and efficiently.Second,research on the super-resolution reconstruction algorithm and propose a new super-resolution reconstruction algorithm in view of the low resolution in screenshots of children's clothing styles,to improve the resolution of photos of children's clothing styles.The latest research on super-resolution reconstruction algorithms only focuses on the quality of image reconstruction,but ignores the importance of image reconstruction time.Since the reconstruction time is too long in the subject research,it will seriously affect the speed of children's clothing style transfer,so it is necessary to establish a light super-resolution reconstruction network not only guarantees the quality of reconstruction,but also reduces the time of reconstruction,to meet the requirements of children's clothing style transfer.Third,research and generate style transfer methods against online children's clothing and respond to the government's " China Smart Manufacturing 2025" strategic strategy of a powerful country to solve the dilemma of the clothing industry's large backlog in terms of no quality in fashion,no fashion in quality;in order to change the conventional clothing design concept,the improved Cyclegan network is used to transfer the overall style and part style of children's clothing styles,and the mutual game learning between the model and the discriminant model is used to resist the generation of network models after super-resolution processing of children's clothing style photos in the era of intelligent design of machine learning,producing a fairly good output of new children's clothing style patterns.
Keywords/Search Tags:Machine learning, Intelligent design, Super-resolution reconstruction, Generative adversarial networks, Style transfer
PDF Full Text Request
Related items