{"title":"A deep learning-based study on visual quality assessment of commercial renovation of Chinese traditional building facades","authors":"Jingjing Zhao , Chenping Han , Yijing Wu , Changsheng Xu , Xing Huang , Xiwu Qi , Yangming Qi , Liang Gao","doi":"10.1016/j.eiar.2025.107862","DOIUrl":null,"url":null,"abstract":"<div><div>In the context of urban renewal, the renovation of Chinese traditional building facades is very important for realizing the renewal of architectural functions and enhancing the urban style. It is of great theoretical and practical significance to assess the visual quality of commercial renovation of Chinese traditional building facades and explore its relationship with the visual features of facades scientifically. Traditional methods based on manual questionnaire surveys and data analysis suffer certain limitations in terms of cost, time, and measurement scale; meanwhile, the research results obtained are prone to be easily influenced by the respondent's subjective preference. In this study, the CRTBFD (Commercial Renovation of Traditional Building Facades Dataset) was firstly constructed, which contained 560 images of commercial renovation of traditional building facades. On account of this dataset, a classification model based on deep learning, Swin-HV (Swin transformer for Historical and cultural atmosphere and Visual preference), was developed. The model can assess and predict the visual quality of commercial renovation of traditional building facades from two aspects: historical and cultural atmosphere and visual preference. In addition, an object detection model called TBFE-YOLO (YOLO for Traditional Building Facade Elements) was proposed, capable of identifying nine traditional building facade elements from reconstructed images. Finally, Spearman's rank correlation coefficient was used to analyze the correlation between visual quality assessment and these nine facade elements, while Grad-CAM++ was employed to further visualize the model's decision-making process. The results show that the Swin-HV model achieves high precision in predicting historical and cultural atmosphere and visual preference assessment. Moreover, it is found that the visual quality assessment of commercial renovation of traditional building facades is closely related to the elements of traditional building facades. The method proposed in this study serves as a reference for urban planning, building conservation, and reuse, while also deepening the understanding of commercial renovation of traditional building facades.</div></div>","PeriodicalId":309,"journal":{"name":"Environmental Impact Assessment Review","volume":"113 ","pages":"Article 107862"},"PeriodicalIF":11.2000,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Environmental Impact Assessment Review","FirstCategoryId":"90","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0195925525000599","RegionNum":1,"RegionCategory":"社会学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENVIRONMENTAL STUDIES","Score":null,"Total":0}
引用次数: 0
Abstract
In the context of urban renewal, the renovation of Chinese traditional building facades is very important for realizing the renewal of architectural functions and enhancing the urban style. It is of great theoretical and practical significance to assess the visual quality of commercial renovation of Chinese traditional building facades and explore its relationship with the visual features of facades scientifically. Traditional methods based on manual questionnaire surveys and data analysis suffer certain limitations in terms of cost, time, and measurement scale; meanwhile, the research results obtained are prone to be easily influenced by the respondent's subjective preference. In this study, the CRTBFD (Commercial Renovation of Traditional Building Facades Dataset) was firstly constructed, which contained 560 images of commercial renovation of traditional building facades. On account of this dataset, a classification model based on deep learning, Swin-HV (Swin transformer for Historical and cultural atmosphere and Visual preference), was developed. The model can assess and predict the visual quality of commercial renovation of traditional building facades from two aspects: historical and cultural atmosphere and visual preference. In addition, an object detection model called TBFE-YOLO (YOLO for Traditional Building Facade Elements) was proposed, capable of identifying nine traditional building facade elements from reconstructed images. Finally, Spearman's rank correlation coefficient was used to analyze the correlation between visual quality assessment and these nine facade elements, while Grad-CAM++ was employed to further visualize the model's decision-making process. The results show that the Swin-HV model achieves high precision in predicting historical and cultural atmosphere and visual preference assessment. Moreover, it is found that the visual quality assessment of commercial renovation of traditional building facades is closely related to the elements of traditional building facades. The method proposed in this study serves as a reference for urban planning, building conservation, and reuse, while also deepening the understanding of commercial renovation of traditional building facades.
在城市更新的背景下,中国传统建筑立面的改造对于实现建筑功能的更新,提升城市风貌具有十分重要的意义。评价中国传统建筑立面商业改造的视觉质量,科学地探讨其与立面视觉特征的关系,具有重要的理论和现实意义。基于手工问卷调查和数据分析的传统方法在成本、时间和测量规模等方面存在一定的局限性;同时,得到的研究结果容易受到被调查者主观偏好的影响。本文首先构建了传统建筑立面商业改造数据集(CRTBFD),该数据集包含560幅传统建筑立面商业改造图像。基于该数据集,开发了基于深度学习的swun - hv(历史文化氛围与视觉偏好的Swin变压器)分类模型。该模型可以从历史文化氛围和视觉偏好两个方面对传统建筑立面商业改造的视觉质量进行评价和预测。此外,提出了TBFE-YOLO (YOLO for Traditional Building Facade Elements)目标检测模型,能够从重建图像中识别出9种传统建筑立面元素。最后,利用Spearman等级相关系数分析视觉质量评价与这9个立面元素之间的相关性,并利用grad-cam++进一步可视化模型的决策过程。结果表明,swan - hv模型在历史文化氛围预测和视觉偏好评价方面具有较高的精度。此外,还发现传统建筑立面商业改造的视觉质量评价与传统建筑立面的要素密切相关。本研究提出的方法可为城市规划、建筑保护和再利用提供参考,同时也加深了对传统建筑立面商业改造的认识。
期刊介绍:
Environmental Impact Assessment Review is an interdisciplinary journal that serves a global audience of practitioners, policymakers, and academics involved in assessing the environmental impact of policies, projects, processes, and products. The journal focuses on innovative theory and practice in environmental impact assessment (EIA). Papers are expected to present innovative ideas, be topical, and coherent. The journal emphasizes concepts, methods, techniques, approaches, and systems related to EIA theory and practice.