中国城市林业2025，Vol.23Issue(4)：61-71,11.DOI:10.12169/zgcsly.2025.03.11.0002
基于街景图像的南京市市区美景度评估及驱动机制时空演变
Spatiotemporal Evolution of Scenic Beauty Estimation and Its Driving Mechanisms in Urban Area of Nanjing Based on Street View Images
摘要
Abstract
[Objective]This study aims to estimate the scenic beauty(SB)of urban street in Nanjing in a scientific and intelligent way and investigate its driving mechanisms,thereby offering theoretical guidance for future street design and management.[Method]Street view images from 2014 to 2022 are collected through web scraping,and visual elements are extracted using the DeepLabV3+model.A ResNet50-based model is constructed to train and predict SB in different areas and analyze their spatiotemporal dynamics.Four driving factors are constructed,including Green View Index(GVI),Sky Openness Index(SOI),Interface Enclosure Index(IEI),and Motor Vehicle Richness Index(MVRI),and the driving mechanisms of SB on the global and local scales and their characteristics are examined using XGBoost-SHAP and Geographically Weighted Random Forest(GWRF)model,respectively.[Result]1)Areas with very low and relatively low SB areas have declined in size over time,while the areas with other levels of SB have generally increased;2)GVI is the most influential factor to SB in 2014-2022.IEI and SOI emerge as dominant factors in 2016 and 2022,respectively,while MVRI consistently makes the weakest contribution;3)GVI exerts a positive influence on SB,whereas SOI and IEI show mixed but generally negative effects,and MVRI exhibit both positive and negative influences;and 4)GVI has the broadest spatial impact in 2014,2015,2019 and 2021,while the influence of other factors varies temporally.[Conclusion]SB in Nanjing city core has improved over time,with GVI identified as the key driver.The SB driving mechanisms demonstrate significant spatial and temporal complexity,and the combination of XGBoost-SHAP and GWRF models proves effective in enhancing the SB interpretability at both global and local levels.关键词
美景度/街景图像/DeepLabV3+/XGBoost-SHAP/地理加权随机森林Key words
scenic beauty/street view image/DeepLabV3+/XGBoost-SHAP/geographically weighted random forest引用本文复制引用
许沉风,李敏,胡一可,耿星,冯紫若,程岩,雒腾宇..基于街景图像的南京市市区美景度评估及驱动机制时空演变[J].中国城市林业,2025,23(4):61-71,11.基金项目
国家自然科学基金重点项目(52038007) （52038007）
