CN 11-5366/S     ISSN 1673-1530
引用本文: 陈星汉,于瀚婷,熊若璟,叶宇.基于空间句法与机器学习的中国古典园林空间指征分析框架建构[J].风景园林,2024,31(3):123-131.
CHEN X H, YU H T, XIONG R J, YE Y. Construction of an Analytical Framework for Spatial Indicator of Chinese Classical Gardens Based on Space Syntax and Machine Learning[J]. Landscape Architecture, 2024, 31(3): 123-131.
Citation: CHEN X H, YU H T, XIONG R J, YE Y. Construction of an Analytical Framework for Spatial Indicator of Chinese Classical Gardens Based on Space Syntax and Machine Learning[J]. Landscape Architecture, 2024, 31(3): 123-131.


Construction of an Analytical Framework for Spatial Indicator of Chinese Classical Gardens Based on Space Syntax and Machine Learning

  • 摘要:
    目的 中国古典园林空间一直以来难以被量化测度,空间句法的兴起使得相关研究向定量化发展,但既有研究与经典理论的融合度仍显不够,且空间特征分析的系统性有待加强。有必要提出一套系统的空间指征分析框架,以支持对古典园林空间的量化测度。
    方法 对中国古典园林空间研究的经典理论进行归纳,使用DepthmapX对园林空间的可视层、可行层模型的各项视域分析指标进行计算,通过叠加分析对空间指征进行测度,借助DBSCAN算法实现对各空间指征聚类特征的识别。以留园、拙政园为例进行分析,并开展感知试验以验证其科学性。
    结果 提出了兼顾人本感知和可测度的5项空间指征:渗透性、曲折度、可视性、可达性和差异度。空间指征的分析框架得到了案例研究与感知试验的支持。
    结论 搭建了一套可操作、易推广的能够系统地提取、刻画并解释古典园林空间特色的指征分析框架,实现了量化分析工具和经典理论的深度融合,探索了中国古典园林空间量化研究的新可能。


    Objective Existing explorations of Chinese classical gardens have predominantly relied on qualitative analysis. The emergence of quantitative research using space syntax has offered a solution to this issue. Although the applicability of space syntax has been validated with eye-level and knee-level application patterns being explored, most of such applications lack systematic and in-depth analysis, thereby limiting their practical effectiveness. To address these issues, this research aims to integrate classical theory with quantitative methods. By extracting spatial indicators from classical theories, this research constructs a framework using quantitative methods such as space syntax and machine learning. Starting from the human perceptual level, this research aims to systematically measure the spatial indicators of classical gardens that were previously deemed “immeasurable”, while also supporting design practice.
    Methods Theoretical framework: This research utilizes classic theoretical works to extract and summarize five spatial indicators that characterize the uniqueness and perceptibility of garden space: permeability, curvature, visibility, accessibility, and differentiation. To incorporate quantitative analysis, a multi-indicator overlay technique is employed to combine spatial indicators, spatial perception experience, and the meaning of visibility graph analysis (VGA) indicators. This approach presents a mapping framework that links indicators to spatial perception and VGA indices. Technical methods: To measure the indicators, VGA is combined with DepthmapX for research on the eye-level and knee-level models of garden space, which involves the calculation of such VGA parameters as connectivity, visual step depth, and integration. ArcGIS is then used to normalize different VGA index values and perform multi-model and multi-index superposition analysis concerning the mapping framework for spatial indicators, achieving preliminary measurement of spatial indicators. The density-based spatial clustering of applications with noise (DBSCAN) algorithm is employed for cluster analysis on the data obtained after spatial superposition analysis in ArcGIS, accurately identifying typical spaces under different dimensions and saliency levels. Liu Garden and the Humble Administrator’s Garden are chosen as measurement examples. To verify the scientificity of the aforesaid measurement methods, a small-scale spatial perception experiment is conducted, employing image questionnaires and perception heat maps to verify the consistency between the analysis results and actual human perception.
    Results Permeability characterizes the spatial hierarchy of the gardens, and the high permeability areas in both gardens exhibit a scattered distribution. Curvature indicates the degree of spatial angle change, with the high curvature areas primarily located along the periphery of both gardens. Spaces with high visibility often take the form of sightseeing corridors, with Liu Garden forming a continuous high visibility channel from southwest to northeast, and visibility in the Humble Administrator’s Garden gradually increasing from south to north. The spatial differentiation in accessibility of Liu Garden diverges linearly to both sides, while that of the Humble Administrator’s Garden diverges from a central point. Differentiation reflects the misalignment of visual and moving lines in space. In Liu Garden, the spaces with high differentiation are mainly found in small-scale courtyards that rely on corridors and windows to create spatial interest. The differentiation cluster in the Humble Administrator’s Garden exhibits a scattered distribution, primarily located in the southern part. Additionally, this research analyzes the spatial characteristics of each level with Liu Garden as an example. Finally, through spatial perception experiments, the research reveals a high consistency between the results of quantified measurement and the subjective perceptions of the participants, providing preliminary evidence for the scientificity and rationality of the spatial indicators extracted and the analytical framework constructed for such spatial indicators.
    Conclusion This research proposes a systematic spatial indicator framework. Compared to previous analyses, the new system more fully depicts the spatial characteristics of classical gardens. Additionally, this research combines various quantitative techniques to establish a deep and easily applicable measurement framework. With this method, a large-scale and refined measurement of existing representative Chinese classical gardens can be quickly achieved, deepening our understanding of the spatial art of classical gardens from a quantitative perspective and promoting the scientific and refined development of the research on spaces of Chinese classical gardens. Moreover, this research effectively empowers design practices by enabling a rapid quantitative evaluation of spatial layouts, organizational structures, and perceptual experience in existing design proposals. Finally, this research comprehensively applies various quantitative methods such as space syntax and machine learning, thereby expanding the technical and methodological resources for quantitative research on garden spaces.