首页|期刊导航|北京建筑大学学报|生成式AI驱动的海南骑楼建筑修复:技术路径与导则构建

生成式AI驱动的海南骑楼建筑修复:技术路径与导则构建OA

Generative AI-Driven Restoration of Hainan Qilou Architecture:Technical Pathways and Guideline Formulation

中文摘要英文摘要

针对海南骑楼因自然老化、不当修缮及人为破坏导致的立面风貌失真与文化原真性流失问题,提出一种基于生成式AI的修复方法.通过构建多模态历史数据库,训练融合建筑知识图谱的LoRA模型,并利用Stable Diffusion实现立面装饰系统破损区域的智能识别与原貌重建.该框架克服了传统修复主观性强、效率低的局限,为文物保护和设计部门提供数据驱动的智能决策支持,有助于科学延续骑楼的文化与气候适应性特征,为建筑遗产可持续保护提供新路径.

To address the degradation of architectural character and cultural authenticity in Hainan Qilou buildings caused by natural aging,inappropriate interventions,and human damage,a generative AI-assisted restoration approach is proposed.A generative AI-based restoration approach is proposed,which involves constructing a multimodal historical database,training a LoRA model integrated with an architectural knowledge graph,and employing Stable Diffusion for intelligent damage detection and authentic appearance reconstruction.This framework overcomes the limitations of traditional restoration methods,such as strong subjectivity and low efficiency,thereby providing data-driven intelligent decision-making support for cultural heritage conservation and design departments.It contributes to the scientifically-grounded preservation of the cultural and climate-adaptive characteristics of Qilou architecture,offering a novel pathway for the sustainable conservation of architectural heritage.

万莹莹;王韬;李花旭

北京建筑大学 建筑与城市规划学院,北京 100044北京建筑大学 建筑与城市规划学院,北京 100044北京建筑大学 建筑与城市规划学院,北京 100044

建筑与水利

海南骑楼建筑遗产修复生成式人工智能多模态数据驱动智能决策

Hainan Qilouarchitectural heritage restorationgenerative artificial intelligencemultimodal data-drivenintelligent decision-making

《北京建筑大学学报》 2026 (2)

61-70,10

北京建筑大学研究生创新项目(PG2024007)北京建筑大学博士研究生科研能力提升项目(DG2025003).

10.19740/j.2096-9872.2026.02.07

评论