基于计算机视觉与街景图像的城市街道 绿化泛类结构量化分析与分布机制研究

Quantitative Analysis and Distribution Mechanismof Urban Street Greening General Structure Basedon Computer Vision and Street View Images

关键词:  风景园林  城市街道绿化泛类结构  街道空间  计 算机视觉  语义分割  Robust回归

Key words:  landscape architecture  urban street greening general structure  street space  computer vision  semantic segmentation  Robust Regression

摘要:

在全球城市化和环境压力加剧的背景下，对城 市街道绿化泛类结构(urban street greening general structure，USGGS)的量化是加强城市区域碳汇、缓解城市 热岛效应以应对全球气候变化的重要前提。通过量化与分析 不同城市的USGGS，探究其与城市建成环境之间的关系。使 用改进的DeepLabV3+神经网络模型，对天津、杭州、深圳 的城市全景街景图像进行语义分割，并结合细粒度数据量化 USGGS，使用Robust回归模型分析USGGS与城市功能属性 POI的关系。研究显示，天津的USGGS主要由单乔木和乔-灌 结构组成，与商业属性和生活属性的POI紧密相关；而杭州和 深圳则展现出包括草本植物在内的多样化USGGS与休闲文化 设施的POI更强的关联性。通过对3个城市USGGS的量化、分 析与比较，为城市绿色基础设施规划和管理奠定了一定的数据 基础，同时基于城市街景图像对USGGS的分析也为城市碳汇 计算与城市热环境研究提供了新的视角。

Abstract:

In the context of global urbanization and increasing environmental pressures, the quantification of Urban Street Greening General Structure (USGGS) is an important prerequisite for enhancing carbon sinks in urban areas and mitigating the urban heat island effect in response to global climate change. By quantifying and analyzing the USGGS in different cities, the relationship between the USGGS and the built environment was explored. The improved DeepLabV3+ neural network model is used to semantically segment the urban panoramic street view images of Tianjin, Hangzhou, and Shenzhen, and combined with the fine-grained data to quantify the USGGS, and the Robust regression model is used to analyze the relationship between the USGGS and the city's functional attributes of POIs. The study shows that Tianjin's USGGS, which mainly consists of single trees and treeshrub structures, is strongly correlated with the POIs of commercial attributes and living attributes, while Hangzhou and Shenzhen show a stronger correlation between diverse USGGS including herbaceous plants and the POIs of leisure and cultural facilities. By quantifying, analyzing and comparing the USGGS of the three cities, a certain data foundation is laid for urban green infrastructure planning and management, and the analysis of the USGGS based on urban streetscape images also provides a new perspective for urban carbon sink calculation and urban thermal environment research.