| 摘要: |
| 蓝绿空间作为城市建成环境的重要组
成部分，不仅能够提升人居环境质量，且对于居
民的身体健康及情绪改善具有重要意义。因此，
如何对蓝绿空间进行规划设计和管理，已成为
间接改善公共健康的关键问题。为了更好地发
掘现有蓝绿空间规划中存在的问题和挑战，优
化蓝绿空间的分布格局，以河北省邯郸市主城
区为例，选取卫星遥感影像构建城市土地利用
分类数据集，基于EfficientNetV2网络模型提取
研究范围内蓝绿空间的位置分布信息，在此基
础上分别计算街区蓝色空间可达性及绿色空间
面积占比，对这两类指标进行归一化加权得到蓝绿空间质量综合评价结果，通过空间自相关探究街区蓝绿空间质量在空间格局上的分异特征。
研究结果表明：1）蓝色空间可达性分布呈现出由内向外递减的趋势，复兴区街区的蓝色空间可达
性相对较低；2）绿色空间网格分布整体呈现出不均衡的状态，绿色空间面积占比相对较低的街
区主要集中在复兴区的南部和邯山区；3）蓝绿空间质量较差的街区有59个，主要分布在复兴区、
丛台区、邯山区的北部。研究采用的卫星遥感影像是免费开源的RGB三通道民用卫星遥感影像，
EfficientNetV2模型用于城市用地分类具有良好的准确率，对于城市建成环境评价具有较强的实
用性。最后根据蓝绿空间质量的量化结果提出系列针对性提升策略，为街区更新的优先选择和方
案制定提供了理论依据。 |
| 关键词: 蓝绿空间 用地分类 EfficientNetV2 蓝色空间可达性 绿色空间面积占比 |
| DOI：10.13791/j.cnki.hsfwest.20240514 |
| 分类号: |
| 基金项目:2023年度河北省社会科学发展研究课题（2023
0203044） |
|
| Evaluation of blue green space quality in urban blocks based on EfficientNetV2 modelland classification: A case study of the urban district of Handan City |
|
WANG Zhenbao,TIAN Yan,ZHANG Xiaoxian,CUI Yidan,LIANG Yuqi
|
| Abstract: |
| As an important component of the urban built environment, blue-green space is of great
significance for improving the living environment and strengthening urban ecological construction.
In addition, increasing scientific evidence shows that blue-green space can provide a variety of
benefits to promote the physical and mental health of residents. They can not only reduce the risk of
chronic diseases such as cardiovascular and respiratory disease but also promote social interaction
and regulate negative emotions. The existing studies evaluate the quality of blue-green space mainly
by calculating the accessibility from streets, communities, and residential areas to large parks, but
lack block scale indicator evaluation, or using street view image data to calculate the street green
and blue visibility rates, without considering the green spaces and water bodies inside parks and
residential areas, which can not reflect the blue-green space quality more comprehensively and
truly. Or the normalized difference vegetation index and the normalized difference water index are
calculated based on satellite remote sensing images, but these two types of indexes cannot clearly
define the land cover type and water body ranges.
In recent years, image classification technology based on deep convolutional neural networks
has developed rapidly, which makes it possible to achieve rapid and accurate land interpretation for
massive remote sensing satellite images in a large area. At present, most datasets used for land use
scenario classification abroad are classified based on their land use characteristics, which cannot
reflect the land classification status of urban built-up area in China well. The existing domestic land
use classification data set, such as the EULUC-China dataset, divides urban built-up area land into
industrial land, public management and public service land, commercial service facilities land and so
on. However, it is difficult to apply to the refined assessment of urban built environment
To better explore the problems in existing blue-green space planning and optimize the
distribution pattern of blue-green space, this paper designs a technical framework for evaluating the
quality of blue-green space in urban blocks based on image classification technology in the field ofdeep learning. First, it selectd satellite remote sensing images to build an urban land use classification dataset. Then, it uses the EfficientNetV2 network model
to quickly and accurately classify and infer land use from satellite remote sensing images of the main urban area of Handan City, extracting the location
distribution information of the blue-green space within the research scope. Using the neighborhood as a research unit, it measures the quality of blue-green
space based on availability and accessibility. The calculation of blue space quality is mainly achieved by creating 64 m x 64 m fishing nets in ArcGIS, using
the Amap API to obtain the shortest walking distance from the center point of each grid in the block to the surrounding water body, and finally, the average
blue space accessibility value of each block is summarized and counted. The calculation of green space quality mainly involves calculating the average green
space area ratio of all grids in the block. On this basis, these two types of indicators are normalized and weighted to obtain the comprehensive evaluation
results of blue-green space quality. Using univariate spatial autocorrelation analysis to explore the differentiation characteristics of blue-green space quality
in the spatial pattern of blocks, and identifying blocks with poor blue-green space quality.
The research results show that: 1) the distribution of blue space accessibility decreases from inside to outside, and the blocks in Fuxing District have
relatively lower blue space accessibility. 2) The distribution of green space is generally uneven, and blocks with a relatively low proportion of green space
area are mainly concentrated in the south of Fuxing District and Hanshan District. 3) There are 59 blocks with poor quality of blue-green space mainly
distributed in the north of Fuxing District, Congtai District, and Hanshan District. Finally, potential grids that can improve the quality of blue-green space
are found in the low-low and low-high clustering blocks. Based on the above research results and these potential grids, a series of improvement strategies are
proposed, including reasonable increment, optimization of inventory, overall planning and system construction.
The satellite remote sensing image data used in the study is open source, free of charge, and easy to obtain, without the need for professional calibration
and preprocessing operations, making it highly practical for urban built environment evaluation. The EfficientNetV2 model has good accuracy in urban land
classification. The refined land classification results can more comprehensively analyze the blue-green space quality at the block scale, helping scientifically
select blocks that need priority renewal. Compared to existing studies that only consider the accessibility of park green space or street green visibility, this article
evaluates the quality of green spaces within parks, residential areas, and streets in a large-scale research area. Based on green spaces, the quality of blue spaces
is also analyzed and discussed. The research framework of this article provides a new approach and method for quantitative analysis of urban built environment. |
| Key words: blue-green space land classification EfficientNetV2 blue space accessibility the proportion of green space area |
