引用本文:
【打印本页】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 92次   下载 398 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于城市街景和深度学习的老龄人群安全感知评价研究 ——以沈阳市中心城区为例
王秋实1, 梁志鹏2, 魏俊添星3, 马雪梅4, 董玉宽5
1.沈阳建筑大学建筑与规划学院,博士研 究生,实验师;2.沈阳建筑大学建筑与规划学院,硕士研 究生;3.沈阳建筑大学建筑与规划学院,硕士 研究生;4.沈阳建筑大学建筑与规划学院,教授;5.( 通讯作者):沈阳建筑大学建筑与规 划学院,教授,博士生导师,dongyk@ sjzu.edu.cn
摘要:
随着我国社会老龄化不断加剧,面向 老龄人群的精细化规划设计方法需求迫切。相 关研究表明,建成环境的感受是否安全,将直接 影响老龄人群开展日常活动的意愿和强度。在 新技术的支持下,大规模快速测度城市空间情绪 感知已成为可能。因此,建立老龄人群的城市空 间“感知—评价—优化”分析框架有助于精细 化规划设计的实现。本研究以重度老龄化的特 大城市——沈阳市中心城区作为研究对象,采用 深度学习对城市街景要素进行语义分割;建立人机对抗的老龄人群安全感知随机森林模型,并对沈阳中心城区的街景进行评价。此外,研究还采 用冷热点聚类、增强归回树(BRT)模型等方法,探讨了城市空间中老龄人群安全感知与建成环 境之间的关系。结果表明:第一,沈阳市老龄安全感知水平呈现显著的空间异质性,总体趋势为内 高外低;第二,研究基于冷热点聚类模型分析得到了安全感知空间聚类特征,并分析其与不同用 地性质的概率分布关系;第三,在不同用地类型下城市老龄安全感知冷热点集聚差异显著,居住 用地热点呈片状分布、商业用地呈南北线状分布、公共服务设施用地由于分散且数量较少而聚类 不显著;第四,研究基于BRT模型分析了人机对抗模型中,街景要素与安全感知水平的贡献水平 和边际效应。研究结果对面向老龄人群的精细化规划和设计提供了一定支撑。
关键词:  深度学习  建成环境  安全感知  老龄友好  城市街景
DOI:10.13791/j.cnki.hsfwest.20240204
分类号:
基金项目:国家自然科学基金面上项目(52178045);辽宁省教 育厅高等学校基本科研项目(LJKMZ20220942、 JYTMS20231584);辽宁省决策咨询和新型智库委 托研究课题(23JC05)
Evaluation of safety perception of elderly people based on urban street view and deeplearning: Taking the central city of Shenyang as an example
WANG Qiushi,LIANG Zhipeng,WEI Juntianxing,MA Xuemei,DONG Yukuan
Abstract:
In recent years, under the background of macro-policy emphasizing on humancentredness, urban design practice and control have gradually transformed from “growth priority” to “quality enhancement”, and the increasing spatial quality of the public has given rise to the study of urban spatial quality from the humanistic perspective. Urban regeneration and reconstruction should not only see “things” but also “people”, and the perception measurement of human scale has become an important basis for urban regeneration decision-making. The health and safety of the elderly in the built environment has always been a common concern of urban planning, urban management, environmental psychology and other disciplines. Safe and comfortable urban space that meets the behavioral characteristics of the elderly population is the basis for improving urban quality. In this context, researchers and designers need to answer the key question of “how the elderly perceive and use space under the human scale, and how and to what extent the physical space affects the perception and behavior of the elderly”. With a series of new technologies, it has become possible to measure the perceived safety level of the aging population from a human-centred perspective in a highly efficient and large-scale manner, and then to achieve a fine-grained planning and design guidance and control. Therefore, the establishment of a “perception-evaluation-optimization” analysis framework of urban space for the aging population is helpful for the realization of fine-tuned planning and design. In this study, the central city of Shenyang, a heavily aging megacity, is taken as the research object, and deep learning is used to semantically segment the elements of the urban streetscape; a random forest model of the safety perception of the aging population with humanmachine confrontation is established, and the streetscape in the central city of Shenyang is evaluated. In addition, the study also used Getis-Ord General G, augmented regression tree (BRT) modeling, and other methods to explore the relationship between the safety perception of the aging population and the built environment in urban space. The results show that, firstly, the level of safety perception of the elderly in Shenyang city shows significant spatial heterogeneity, the overall trend is high inside and outside, and the urban ring road shows a more obvious spatial progressive relationship; secondly,the study based on the analysis of spatial autocorrelation model to obtain the characteristics of the spatial clustering of safety perception, and analyze the probability of its relationship with the distribution of the nature of the different land use; the relationship of high clustering of the perception of safety (high: perception of safety) area with commercial land, residential land The relationship between the high clustering of security perception (high: security perception) areas and commercial land, residential land, public service facilities land is more obvious; the relationship between the low clustering (high: security perception) areas and industrial land, logistics and warehousing land is more obvious; thirdly, the trend of spatial distribution of the level of security perception in the neighborhoods of different land use types is similar to the trend of the overall level of the city, but there are some differences; residential land use is not obvious in the clustering of the city boundary between the new and old urban areas, and the hot spot clustering area extends northward in addition to the old urban area; the hotspot clustering area of commercial land is distributed in two lines in the north and south, with a high degree of overlap between the main metro lines of the city and the nodes of the commercial circle; public service facilities due to the number of parcels of land its dispersion, the cold hotspot clustering is not significant; fourthly, the study analyses, based on the BRT model, the level of the contribution of the elements of the streetscape with the level of security perception in the model of the human-computer confrontation and the marginal effect buildings (interface), sky (visibility), paving and walls, which shape the basic form and enclosure relationships of urban space. This study takes human-scale urban spatial data as a blueprint, and relies on new technologies and large models to carry out fine-grained measurement and analysis of large-scale perception for the aging population. Through data analysis, urban problems are revealed, and problem-oriented urban analysis is efficiently, rapidly and scientifically realized, which helps the science of decision-making. The limitation of the study is that the volunteers of the humancomputer confrontation model for the aging population in this paper come from a relatively single source, and individual differences such as educational background are not taken into account, so we expect that the accuracy of the urban perception evaluation model will be revised in the future.
Key words:  deep learning  built environment  safety perception  age-friendly  urban street view