基于LoRA标签优化的敦煌藻井图案AI生成式设计研究

孟欣怡; 殷晓晨; 赵毅龙

doi:10.19554/j.cnki.1001-3563.2026.02.019

PDF(36062 KB)

包装工程（设计栏目） ›› 2026, Vol. 47 ›› Issue (2) : 197-210. DOI: 10.19554/j.cnki.1001-3563.2026.02.019

视觉传达设计

基于LoRA标签优化的敦煌藻井图案AI生成式设计研究

孟欣怡, 殷晓晨^*, 赵毅龙

作者信息 +

AI Generative Design of Dunhuang Ceiling Patterns Based on LoRA Label Optimization

MENG Xinyi, YIN Xiaochen^*, ZHAO Yilong

Author information +

文章历史 +

摘要

目的将设计学的方法融入AI生成式设计流程,通过提取LoRA训练集特征优化标签内容,增强LoRA风格迁移效果。方法以敦煌藻井图案为例,构建训练集。首先,通过结构、纹样元素分析,色彩聚类和感性工学实验,提取样本既定的图像特征和能概括样本风格的感性词汇;其次,基于眼动追踪技术对藻井图案中纹样元素的特征性进行量化分析,根据视觉显著性对纹样排序;最后,将特征提取结果转换为自然语言,对训练集手动标注。结果对比分析了优化标签文本前后使用LoRA模型的生成内容,结果显示,优化后的模型再现训练集样本风格特征和纹样内容的能力更好。结论通过结合定性和定量方法,能够客观全面提取训练集特征,有效优化标签文本,提高使用LoRA模型生成内容的稳定性,进而通过智能设计推动传统纹样的创新与发展。

Abstract

The work aims to integrate design methodologies into AI generative design processes, and enhance LoRA style transfer effectiveness by optimizing label content through feature extraction from LoRA training sets. With Dunhuang ceiling patterns as an example, a training set was constructed. Firstly, the established image features of the samples and perceptual keywords summarizing the styles were extracted through structural analysis, pattern element analysis, color clustering, and Kansei engineering experiments. Secondly, eye-tracking technology was employed to quantify the characteristic features of pattern elements in the ceiling patterns, ranking them based on visual importance. Finally, the extracted features were converted into natural language for manual annotation of the training set. A comparison of generative content before and after label optimization using the LoRA model demonstrated that the optimized model more effectively reproduced the style characteristics and pattern content of the training set samples. By combining qualitative and quantitative methods, the study objectively and comprehensively extracts training set features, effectively optimizes label content, and improves the stability of content generated by the LoRA model. This approach promotes the innovation and development of traditional patterns through intelligent design.

导出引用1

孟欣怡, 殷晓晨, 赵毅龙. 基于LoRA标签优化的敦煌藻井图案AI生成式设计研究[J]. 包装工程. 2026, 47(2): 197-210 https://doi.org/10.19554/j.cnki.1001-3563.2026.02.019

MENG Xinyi, YIN Xiaochen, ZHAO Yilong. AI Generative Design of Dunhuang Ceiling Patterns Based on LoRA Label Optimization[J]. Packaging Engineering. 2026, 47(2): 197-210 https://doi.org/10.19554/j.cnki.1001-3563.2026.02.019

中图分类号： TB472

参考文献

[1] 卢兆麟, 宋新衡, 金昱成. AIGC技术趋势下智能设计的现状与发展[J]. 包装工程, 2023, 44(24): 18-33.
LU Z L, SONG X H, JIN Y C.State of Arts and Development of Intelligent Design Methods under the AIGC Trend[J]. Packaging Engineering, 2023, 44(24): 18-33.
[2] SHI Y, SHANG M Y, QI Z Q.Intelligent Layout Generation Based on Deep Generative Models: A Comprehensive Survey[J]. Information Fusion, 2023, 100: 101940.
[3] ZHOU C Y, ZHANG X Y, YU C Y.How Does AI Promote Design Iteration? The Optimal Time to Integrate AI into the Design Process[J]. Journal of Engineering Design, 2025, 36(11): 1904-1931.
[4] 王子祥, 贾小军, 冉二飞, 等. 基于改进稳定扩散模型的蓝印花布纹样生成[J]. 光电子·激光, 2025(7): 712-721.
WANG Z X, JIA X J, RAN E F, et al.Blue Calico Pattern Generation Based on Improved Stable Diffusion Model[J]. Journal of Optoelectronics·Laser, 2025(7): 712-721.
[5] 于鹏, 张毅. 从特征辨识到图像生成: 基于AIGC范式的苗族服饰设计[J]. 丝绸, 2024, 61(3): 1-10.
YU P, ZHANG Y.From Feature Recognition to Image Generation: Miao Ethnic Costume Design Based on the AIGC Paradigm[J]. Silk, 2024, 61(3): 1-10.
[6] 李自强, 杨薇, 杨先凤, 等. 基于弱标签争议的半自动分类数据标注方法[J]. 电子学报, 2024, 52(8): 2891-2899.
LI Z Q, YANG W, YANG X F, et al.The Semi-Automatic Classification Data Labeling Method Based on Dispute about Weak Label[J]. Acta Electronica Sinica, 2024, 52(8): 2891-2899.
[7] KULKARNI G, PREMRAJ V, ORDONEZ V, et al.BabyTalk: Understanding and Generating Simple Image Descriptions[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(12): 2891-2903.
[8] CHEN F H, JI R R, SUN X S, et al.GroupCap: Group-Based Image Captioning with Structured Relevance and Diversity Constraints[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018.
[9] KARPATHY A, JOULIN A, FEI-FEI L F. Deep Fragment Embeddings for Bidirectional Image Sentence Mapping[J]. Advances in Neural Information Processing Systems, 2014, 27: 5679.
[10] 常致富, 周风余, 王玉刚, 等. 基于深度学习的图像自动标注方法综述[J]. 山东大学学报(工学版), 2019, 49(6): 25-35.
CHANG Z F, ZHOU F Y, WANG Y G, et al.A Survey of Image Captioning Methods Based on Deep Learning[J]. Journal of Shandong University (Engineering Science), 2019, 49(6): 25-35.
[11] 杨诗曼, 王中训, 吴文静, 等. 图像自动标注技术研究[J/OL]. 探测与控制学报, 2025: 1-12. (2025-02-27) [2025-03-19]. http://kns.cnki.net/kcms/detail/61.1316.TJ.20240724.1138.006.html.
YANG S M, WANG Z X, WU W J, et al. Research on Automatic Image Annotation Technology[J/OL]. Journal of Detection & Control, 2025: 1-12. (2025-02-27) [2025- 03-19]. http://kns.cnki.net/kcms/detail/61.1316.TJ.20240724.1138.006.html.
[12] DHARIWAL P, NICHOL A. Diffusion Models Beat GANs on Image Synthesis[EB/OL].2021: arXiv: 2105. 05233. https://arxiv.org/abs/2105.05233
[13] ROMBACH R, BLATTMANN A, LORENZ D, et al.High-Resolution Image Synthesis with Latent Diffusion Models[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans: IEEE, 2022.
[14] HU E J, SHEN Y L, WALLIS P, et al. LoRA: Low-Rank Adaptation of Large Language Models[EB/OL].2021: arXiv: 2106.09685. https://arxiv.org/abs/2106.09685.
[15] SUNDAR S S.Rise of Machine Agency: A Framework for Studying the Psychology of Human-AI Interaction (HAII)[J]. Journal of Computer-Mediated Communication, 2020, 25(1): 74-88.
[16] 宋筱璇, 徐丹, 张紫蕴, 等. 提示词重构类型对AI生成内容效果的影响[J]. 图书馆论坛, 2025(4): 129-138.
SONG X X, XU D, ZHANG Z Y, et al.Research on the Influence of Prompt Reformulation Types on Performance of Artificial Intelligence Generated Content[J]. Library Tribune, 2025(4): 129-138.
[17] 路紫媚. 敦煌壁画中的卷草纹样研究[D]. 北京: 北京林业大学, 2021.
LU Z M.Research on the Pattern of Curly Grass in Dunhuang Frescoes[D]. Beijing: Beijing Forestry University, 2021.
[18] 王乙婷. 敦煌西夏时期藻井图案的装饰性及包装设计应用研究[D]. 沈阳: 鲁迅美术学院, 2023.
WANG Y T.A Study on the Decorative and Packaging Design Application of Diatoma Well Patterns during the Western Xia Dynasty in Dunhuang[D]. Shenyang: Luxun Academy of Fine Arts, 2023.
[19] 陈振旺, 郭美娟, 王爱婷. 隋代中期莫高窟藻井图案研究[J]. 敦煌研究, 2020(1): 51-59.
CHEN Z W, GUO M J, WANG A T.A Study on the Central Ceiling Patterns in the Caves at Mogao from the Middle Tang Period[J]. Dunhuang Research, 2020(1): 51-59.
[20] 卜俊, 孙培贤, 唐刚, 等. 敦煌藻井纹样在现代女式手提包中的创新设计与应用[J]. 皮革科学与工程, 2021, 31(3): 71-76.
BU J, SUN P X, TANG G, et al.The Innovative Design and Application of Dunhuang Caisson Patterns in Modern Female Handbag[J]. Leather Science and Engineering, 2021, 31(3): 71-76.
[21] 余兰. 敦煌藻井纹样在服饰设计中的数字化创新应用[J]. 毛纺科技, 2022, 50(9): 45-53.
YU L.Digital Innovation Applications of Dunhuang Caisson Pattern in Costume Design[J]. Wool Textile Journal, 2022, 50(9): 45-53.
[22] 刘悦. 面向敦煌藻井的纹样生成设计方法研究[D]. 兰州: 兰州理工大学, 2023.
LIU Y.Research on Design Methods for Pattern Generation for Dunhuang Algal Well[D]. Lanzhou: Lanzhou University of Technology, 2023.
[23] 陈鋆纯, 彭坚, 郭寅曼, 等. 基于花瑶挑花纹样风格化图像的辅助设计研究[J]. 包装工程, 2022, 43(14): 246-253.
CHEN (J /Y)C, PENG J, GUO Y M, et al. Computer-Aided Design Based on Stylized Images of Huayao Cross-Stitch Patterns[J]. Packaging Engineering, 2022, 43(14): 246-253.
[24] 常沙娜. 中国敦煌历代装饰图案[M]. 北京: 清华大学出版社, 2009: 47-64.
CHANG S N.Decorative Design from China Dunhuang Murals[M]. Beijing: Tsinghua University Press, 2009: 47-64.
[25] 杨东苗. 敦煌历代精品藻井100图[M]. 杭州: 浙江古籍出版社, 2010.
YANG D M.100 Pictures of Fine Algae Wells in Dunhuang in Past Dynasties[M]. Hangzhou: Zhejiang Ancient Books Publishing House, 2010.
[26] 段媛媛. 试论敦煌莫高窟十六国至北朝时期覆斗形顶(上)——莫高窟覆斗顶与晋墓顶[J]. 敦煌研究, 2021(2): 85-96.
DUAN Y Y.A Study on the Truncated Pyramidal Ceilings of Mogao Caves during the Sixteen Kingdoms Period and Northern Dynasties(Ⅰ)—The Truncated Pyramidal Ceilings of Mogao Caves and the Jin Dynasty Tomb Ceilings[J]. Dunhuang Research, 2021(2): 85-96.
[27] 杨梅, 李劲松, 王怡妍. 敦煌传统壁画色彩网络模型构建与应用设计[J]. 包装工程, 2020, 41(18): 222-228.
YANG M, LI J S, WANG Y Y.Construction and Application of Color Network Model of Dunhuang Traditional Fresco[J]. Packaging Engineering, 2020, 41(18): 222-228.
[28] KOBAYASHI S.The Aim and Method of the Color Image Scale[J]. Color Research & Application, 1981, 6(2): 93-107.
[29] 梁张英. 均衡、比例、秩序[D]. 杭州: 中国美术学院, 2022.
LIANG Z Y. Equilibrium, Proportion, Order[D]. Hangzhou: China Academy of Art, 2022.
[30] WU F, HSIAO S W, LU P.An AIGC-Empowered Methodology to Product Color Matching Design[J]. Displays, 2024, 81: 102623.
[31] 李博, 宣金鸽, 薛艳敏, 等. 基于视觉因素的地铁调度人机界面优化设计评价[J]. 图学学报, 2025, 46(1): 200-210.
LI B, XUAN J G, XUE Y M, et al.Evaluation of Optimal Design of Human-machine Interface for Subway Dispatching Based on Visual Factors[J]. Journal of Graphics, 2025, 46(1): 200-210.
[32] 赵新灿, 左洪福, 任勇军. 眼动仪与视线跟踪技术综述[J]. 计算机工程与应用, 2006, 42(12): 118-120.
ZHAO X C, ZUO H F, REN Y J.A Review of Eye Tracker and Eye Tracking Techniques[J]. Computer Engineering and Applications, 2006, 42(12): 118-120.
[33] 苏建宁, 鱼宝银, 李雄, 等. 意象驱动的产品造型智能设计方法[J]. 机械设计, 2024, 41(8): 115-120.
SU J N, YU B Y, LI X, et al.Image-Driven Product Modeling Intelligent Design Methodology[J]. Journal of Machine Design, 2024, 41(8): 115-120.