摘要
汉服做为一种穿着时尚,深受年轻人的喜爱,但现代汉服的风格信息却难以被许多汉服爱好者准确辨识。在YOLOv5s模型的基础上,插入Repvgg模块的同时,引入SE注意力机制来提高模型的网络特征提取能力;使用SIoU_Loss优化损失函数提升边界框定位精度,从而达到实时检测汉服风格的目的。结果表明:该算法明显改善多项评价指标,整体精确率达到92.4%,召回率达到91.6%,平均精度均值达到91.8%,单张图像推理时间仅需15.0 ms。该方法能够快速准确地辨识汉服风格,帮助人们了解现代汉服的风格特征,为中华优秀传统文化的传承发展提供技术支持。
Hanfu has become a kind of wearing fashion for young people,but its modern style information is difficult to be accurately identified by many Hanfu fans.Therefore,based on the YOLOv5s model,this paper improved the network feature extraction capability by inserting the Repvgg module and introducing the SE attention mechanism at the same time,and optimized the loss function using SIoU_Loss to improve the bounding box positioning accuracy,so as to realize the real-time detection of Hanfu style.The experimental results show that the algorithm achieves significant improvement in a number of evaluation indexes,with an overall precision rate of 92.4%,a recall rate of 91.6%,an average precision mean of 91.8%,and a single-image inference time of only 15.0 ms.The method can recognize the style of Hanfu quickly and accurately,thus helping people to understand the stylistic characteristics of modern Hanfu and providing technical support for the inheritance and development of Chinese excellent traditional culture.
作者
张俊杰
蒋博闻
袁桦
李丽
朱强
ZHANG Junjie;JIANG Bowen;YUAN Hua;LI Li;ZHU Qiang(School of Computer Science and Artificial Intelligence,Wuhan Textile University,Wuhan 430200,Hubei,China;Hubei Provincial Engineering Research Center for Intelligent Textile and Fashion,Wuhan 430200,Hubei,China;Engineering Research Center of Hubei Province for Clothing Information,Wuhan 430200,Hubei,China;School of Fashion,Wuhan Textile University,Wuhan 430073,Hubei,China;Wuhan Textile and Apparel Digital Engineering Technology Research Center,Wuhan 430073,Hubei,China)
出处
《北京服装学院学报(自然科学版)》
CAS
2024年第1期95-103,118,共10页
Journal of Beijing Institute of Fashion Technology:Natural Science Edition
基金
科技部重点研究专项“面向纺织服装产业集聚区域的网络协同制造集成技术研究与示范”(2019YFB1706300)
湖北省教育厅哲学社会科学研究重点项目“双碳背景下基于多源异构数据的个性化服装推荐机制研究”(22D062)
纺织服装智能化湖北省工程研究中心2022年度开放课题“基于深度学习的智能服装产品开发与推荐系统设计”(2022HBITF06)。