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

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 optimization,and selected the optimal algorithm for each index for the task of extracting Yunjian ornamentation patterns in Qing Dynasty.In addition,we obtained the segmentation results of complex and simple motifs under this task and analyzed them by extracting the patterns of Yunjian motifs in the Qing Dynasty.We used the semantic segmentation method in deep learning to extract the ornamentation patterns of Yunjian in the Qing Dynasty,and took speediness,loss reduction,accuracy and convenience as the main objectives of the selected algorithm to study the path extraction of ornamentation pattern elements of Yunjian in the Qing Dynasty.The experimental results show that the U-Net algorithm performs optimally on the image dataset of Yunjian in the Qing Dynasty,and can effectively improve the accuracy of the extraction of Yunjian in the Qing Dynasty.The accuracy,recall,average intersection ratio and Dice similarity coefficient of this algorithm reach 97.57%,97.84%,92.04%and 93.27%,respectively,which can achieve accurate automatic ornament extraction and provide an effective and reliable path method for automatic ornament pattern extraction of Yunjian in the Qing Dynasty.The integration of deep learning and Yunjian in the Qing Dynasty can provide enlightenment for the digital protection of traditional fabrics.On the one hand,the use of artificial intelligence technology to extract the Yunjian decoration can effectively enrich the traditional Yunjian decoration database,and play a role in the digital protection of the traditional Yunjian.On the other hand,the digital transformation of Yunjian decoration image can provide new ideas for the inheritance and utilization of traditional Yunjian,simplify the extraction method of decoration,promote the inheritance of Yunjian element culture,and improve the utilization rate of Yunjian elements.The research results can provide a reference for the digital protection and application of traditional Yunjian patterns.