基于深度学习的清代云肩纹饰图案提取路径研究  

Research on the extraction path of the Qing Dynasty Yunjian patterns based on deep learning

作　　者：关瑛[1,2] 于瀚超 王秀峰[3] 曹新强[1] GUAN Ying;YU Hanchao;WANG Xiufeng;CAO Xinqiang(School of Art&Design,Shaanxi University of Science&Technology,Xi’an 710016,China;Silk Road Cultural Heritage Innovation Design Research Center,Shaanxi University of Science&Technology,Xi’an 710016,China;School of Materials Science&Engineering,Shaanxi University of Science&Technology,Xi’an 710016,China)

机构地区：[1]陕西科技大学设计与艺术学院,西安710016 [2]陕西科技大学丝路文化传承与创新设计研究中心,西安710016 [3]陕西科技大学材料科学与工程学院,西安710016

出　　处：《丝绸》2023年第2期123-129,共7页Journal of Silk

基　　金：陕西省社会科学联合会项目(2020Z067);四川省教育厅项目(DM20216Z)。

摘　　要：云肩是传统民族服饰艺术中的重要服饰部件,其纹饰具有民族特色和文化底蕴,文章为提高清代云肩纹饰图案数字化保护的有效性和准确率,建立了语义分割中U-Net网络提取清代云肩纹饰的系统路径。首先,搭建基于U-Net的算法模型对清代云肩纹饰图像进行提取;其次,将U-Net模型训练的关键参数优化对比,选出其最优参数;最后,将训练好的U-Net模型与其他网络模型进行对比实验。实验结果表明,该算法的准确率、召回率、平均交并比和Dice相似系数分别达到97.57%、97.84%、92.04%、93.27%,可以实现精确的纹饰自动提取,具有理想的有效性和可靠性,对清代云肩纹饰图案数字化保护与应用具有良好的应用价值。The Qing Dynasty was the heyday in the history of Yunjian,with rich variety,meticulous workmanship,exquisite decoration,and rich cultural connotation.It is the product of the combination of practical function and display function and is a very distinctive art type in the treasure house of ancient Chinese art.However,as Yunjian is widely spread among the people,its preservation is not fine enough,and data collection is difficult,it is difficult to find samples in the process of studying Yunjian ornament,which seriously affects the research of collectors and scholars,and limits the protection and application of Yunjian ornament.In order to facilitate the improvement of the traditional Yunjian ornament database and the precise integration of science and technology and preservation of cultural relics,we adopted the deep learning theory to solve the problems of Yunjian ornamentation in the Qing Dynasty,such as centralized data,obscure image color and blurred outline edge.By establishing a systematic path of Yunjian ornamentation extraction by U-Net network in semantic segmentation,we obtained complete Yunjian ornamentation patterns in the Qing Dynasty,which effectively improved the effectiveness and accuracy of digital protection of Yunjian ornamentation patterns in the Qing Dynasty.First,we built the U-Net network model.We used this algorithm model to extract the images of Yunjian patterns in the Qing Dynasty.During the training process,we iterated the image data and samples with the real marks of the Qing Dynasty Yunjian in the model.Secondly,we optimized and compared the parameters.We optimized and adjusted the key parameters in the U-Net model,and compared the parameters,and applied the trained U-Net optimal network to the Yunjian ornamentation image in Qing Dynasty for ornamentation segmentation.Finally,we compared the algorithms for experiments.We selected the optimal model of each network to segment this dataset for ornamentation,compared the segmentation results with those of the U-Net model after parameter op

关 键 词：清代云肩 语义分割 U-Net 纹饰提取 参数训练 数字化保护 

分 类 号：TS941.26[轻工技术与工程—服装设计与工程] J523.6[艺术—艺术设计]