[1]陈世婕,王卫星*,彭 莉.基于多尺度网络的苗绣绣片纹样分割算法研究[J].计算机技术与发展,2023,33(11):149-155.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 022]
　CHEN Shi-jie,WANG Wei-xing*,PENG Li.Research on Miao Embroidery Pattern Segmentation Algorithm Based on Multi-scale Network[J].,2023,33(11):149-155.[doi:10. 3969 / j. issn. 1673-629X. 2023. 11. 022]

点击复制

基于多尺度网络的苗绣绣片纹样分割算法研究()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年11期

页码:

149-155

栏目:

人工智能

出版日期:

2023-11-10

文章信息/Info

Title:

Research on Miao Embroidery Pattern Segmentation Algorithm Based on Multi-scale Network

文章编号:

1673-629X(2023)11-0149-07

作者:

陈世婕; 王卫星* ; 彭 莉

贵州大学 机械工程学院,贵州 贵阳 550025

Author(s):

CHEN Shi-jie; WANG Wei-xing* ; PENG Li

School of Mechanical Engineering,Guizhou University,Guiyang 550025,China

关键词:

苗绣; 数据库; 深度学习; 语义分割; 数字化保护

Keywords:

Miao Embroidery; database; deep learning; semantic segmentation; digital protection

分类号:

TP391. 41

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 11. 022

摘要:

苗绣作为苗族人民传承历史文化的主要形式,图案内容丰富。 但纹样设计周期相对较长且大多为手工劳动,因此生产效率较低。 与此同时,在互联网平台上的相关资源质量不佳且

数量较少,大多纹样及实体获取往往需要到实地考察。为解决传统手工苗绣绣片难以保存的问题,采用深度学习的图像语义分割方法进行相关苗绣绣片纹样提取用以保存和分类。?

该文构建了一组小型苗绣绣片纹样数据库并进行相关类别标注,同时提出一种多尺度图像语义分割网络架构( Multi-Scale Network) 。 该方法 包 含 四 个 模 块, 分 别 为 多 尺 度 模 块 ( Multi - Scale Block, MSB) 、 多 尺 度 编 码 器 ( Multi - ScaleEncoder,MSE) 、多尺度解码器( Multi-Scale Decoder,MSD)以及语义分割头( Semantic-Head,SH) 。 每个模块均

采用高效设计的多尺度提取机制和残差结构,可以提取更多全局纹样信息,并显著提升模型在不同任务下的泛化能力。 所提方法在 PASCAL VOC 2012 数据集以及自建苗绣绣片纹样数据库中进行测试,实验结果相比全卷积神经网络的交并比( MIoU)提高了 14.84% 和 6. 7 % 。 在相同实验条件下也验证了算法在应对错误分割、遗漏分割的有效性。

Abstract:

The Miao Embroidery in Qiandongnan,Guizhou Province,China is a precious intangible cultural heritage,as well as a nationalcostume handicraft and textile. Its exquisite pattern design requires exquisite workmanship. However,the design cycle is relatively longand most of them are manual labor,so the production efficiency is low. At the same time,the quality and quantity of relevant resourceson the Internet platform are poor,and most of the patterns and entities need to be obtained on the spot. In order to solve the problem thattraditional manual Miao Embroidery pieces are difficult to save,the image semantic segmentation method based on deep learning is usedfor saving and classification. We have constructed a group of small Miao Embroidery pattern databases and proposed a multi-scale imagesemantic segmentation network architecture ( Multi Scale Network ) . The method includes four modules, namely, Multi Scale Block( MSB) ,Multi Scale Encoder ( MSE) ,Multi Scale Decoder ( MSD) and Semantic Head ( SH) . Each module adopts an efficient multi-scale extraction mechanism and residual structure,which can extract more global information and significantly improve the generalizationof the model under different tasks. The proposed method was tested in PASCAL VOC 2012 dataset and self built Miao Embroidery PiecePattern Database. The experimental results showed that?

the mean intersection over union ( MIoU) of the full convolution neural networkwas improved by 14. 84% and 6. 7% compared with the full convolution neural network. Under the same experimental conditions,theproposed method in dealing with error segmentation and omission segmentation is also verified.

常用功能