[1]陈 桔,王芳宇,郑 溪,等.城市空间特色视觉感知测度与影响因素分析——以昆明市地铁站口为例[J].中国名城,2025,39(6):67-75.[doi:10.19924/j.cnki.1674-4144.2025.006.010]
　CHEN Ju,WANG Fangyu,ZHENG Xi,et al.Visual Perception Measurement and Influential Factor Analysis of Urban Spatial Features: A Case Study of Subway Station Entrances in Kunming City[J].China Ancient City,2025,39(6):67-75.[doi:10.19924/j.cnki.1674-4144.2025.006.010]
点击复制

城市空间特色视觉感知测度与影响因素分析——以昆明市地铁站口为例()
分享到：

中国名城[ISSN:1674-4144/CN:32-1793/GO]

卷:
39
期数:
2025年6期
页码:
67-75
栏目:
出版日期:
2025-06-05

文章信息/Info

Title:
Visual Perception Measurement and Influential Factor Analysis of Urban Spatial Features: A Case Study of Subway Station Entrances in Kunming City
文章编号:
1674-4144（2025）006-0067-09
作者:
陈 桔; 王芳宇; 郑 溪; 胡博文; 李 朗
陈 桔，昆明理工大学建筑与城市规划学院副教授；王芳宇，昆明理工大学建筑与城市规划学院硕士研究生；郑 溪，昆明理工大学建筑与城市规划学院博士研究生；胡博文，南京市城市与交通规划设计研究院股份有限公司工程师；李 朗，昆明理工大学建筑与城市规划学院硕士研究生。
Author(s):
CHEN Ju; WANG Fangyu; ZHENG Xi; HU Bowen; LI Lang
关键词:
街景图像; 可解释机器学习; 城市空间特色; 地铁站口; 昆明
Keywords:
street images; interpretable machine learning; urban spatial features; subway station entrances; Kunming City
分类号:
TU984
DOI:
10.19924/j.cnki.1674-4144.2025.006.010
文献标志码:
A
摘要:
以昆明市地铁站口为研究对象，构建城市空间特色视觉感知的定量测度模型，探讨城市重要公共节点空间中影响空间特色视觉感知的关键因素。首先，采集昆明市近300个地铁站口的街景图像，结合光学字符识别（OCR）技术与PSPNet语义分割算法提取建成环境要素。然后，基于网络游记文本识别“民族风情”“历史风貌”及“城市氛围”3类空间特色标签，邀请市民进行图像可感知性评分。最后，采用随机森林与Extra Trees等机器学习算法进行模型训练，并利用SHAP（SHapley Additive exPlanations）方法分析各建成环境要素对感知评分的影响。研究发现：昆明市地铁站口的城市空间特色感知水平在二环路内整体较高，历史风貌高感知水平站点集中在“小三山一水”格局核心区域，城市氛围则呈多核心分布，民族风情在站口侧向视角表现更佳；SHAP分析揭示不同建成环境要素对感知的差异化影响，例如民族风情感知受地面铺装及街景色彩丰富度等要素正向促进，而过多树木阻碍感知。以期为机器学习在城市特色视觉感知领域的应用提供新的视角，并为中微观层面城市空间特色规划的精细化实施提供理论依据。
Abstract:
This study focuses on the subway station entrances in Kunming City, constructing a quantitative model for visual perception of urban spatial features and exploring the key factors influencing the perception of spatial characteristics in these important public node spaces. First, street view images of 300 subway station entrances in Kunming are collected, and built environment elements are extracted through OCR technology and the PSPNet semantic segmentation algorithm. Next, three spatial feature labels— "ethnic characteristics," "historical features," and "urban atmosphere"—are identified based on online travelogue texts, and citizens are invited to rate the perceptibility of these features in the images, thereby constructing a training dataset. Machine learning algorithms, such as Random Forest and Extra Trees, are used for model training, and SHAP methods were applied to analyze the impact of various built environment elements on the perception scores. The findings reveal that the overall perception level of urban spatial features at Kunming’s subway station entrances is higher within the Second Road, that the sites with high perceptibility of historical features are concentrated in the core area of the "Small Three Mountains and One Water" urban pattern, and that the perception of urban atmosphere exhibits a multi-core distribution, while ethnic characteristics are better perceived from side-view perspectives of the station entrances. SHAP analysis reveals the differentiated impact of various built environment elements on perception. For example, the perception of "ethnic characteristics" is positively influenced by elements such as ground paving and streetscape color richness, whereas an excessive number of trees can impede this perception. This study provides a new perspective on the application of machine learning in the field of urban feature visual perception and offers a theoretical basis for the refined implementation of urban spatial feature planning at the micro and meso-scale.
更新日期/Last Update: 2025-06-17