| 摘要: |
| 随着数字化和人工智能发展，机器
学习在乡村景观空间分析和规划中应用前景
广阔。文章基于Web of Science 数据库的108
篇文献，分析机器学习在乡村景观空间分析
中的应用。研究主题分为乡村空间形态、环
境和社会经济文化三大类，其中乡村空间形
态涵盖空间布局、公共空间和建筑空间；环
境包括土地利用（农业景观、水环境、植物
群落）、气候及危害风险；社会经济文化涉
及社会生态系统、文化景观和经济发展与环
境关系。卷积神经网络、随机森林和支持向
量机是最受欢迎的机器学习方法，卫星图
像、土地覆盖和数字高程模型数据是常用基
础数据。案例主要分布于亚洲、非洲和欧
洲，中美洲、南美洲和大洋洲较少。文献数
量增长表明该领域发展迅速。研究指出当前
应用进展、研究空白及未来发展趋势。 |
| 关键词: 机器学习 综述 乡村景观 地理
空间数据 风景园林 |
| DOI：10.13791/j.cnki.hsfwest. 20240329002 |
| 分类号: |
| 基金项目:中建五局科技研发课题（cscec5b-2023-01） |
|
| Application and prospect of machine learning in spatial analyses of rural landscape abroad |
|
HE Qingying,YANG Ying,GUO Taoyang
|
| Abstract: |
| With the development of digitalization and the maturity of Artificial Intelligence (AI)
technologies, big data and geospatial data services, machine learning is expected to bring innovation
to spatial analyses and planning tools for rural landscapes. The article mainly focuses on illustrating
the most frequent topics, the most popular machine learning algorithms, and the common types of
geospatial data used in the spatial analysis of rural landscapes, systematically summarizing machine
learning methods in spatial analysis of rural landscapes. This study has settled up literature screening
criteria, using the Web of Science (WOS) as the primary resource and Google Scholar as a
complementary database, in order to obtain more valuable literatures. The timeframe of literature
publication covered the period from 2012 to April 2023, and the selected studies focused on applying
machine learning to solve problems in rural landscapes as well as using geospatial data to obtain and
analyze relevant information. Finally, 108 papers meeting our criteria were selected using the Web of
Science (WOS) as the premier database. After screening the literature, they were analyzed in detail.
Key information extracted included title, keywords, publication date, purpose of the study, study area,
machine learning methods used and geospatial data. By combing and analyzing them, the conclusions
are as follows. Firstly, the research themes were classified into 3 major categories and 9
subcategories, the 3 major categories were rural spatial patterns, environment and socio-economic
culture, and the 9 subcategories were the refinement classification of the 3 major categories. It could
be able to identify themes that already exist in current research with the thorough literature reviews
based on the classification. Rural spatial patterns were divided into rural spatial layouts, rural public
spaces, and rural architectural spaces. Besides, the environment aspect was divided into land use,
climate, and rural area hazards and risks. Specifically, land use covers agricultural landscapes, water
environments, and plant communities, hazards and risks encompass disaster risk and pollution. Last,
the socio-economic cultural perspective contains social-ecological systems, cultural landscapes, and
the relationship between economic development and the environment. Among these, the study of
social-ecological systems focuses on the cultural services of ecosystems and coupled socio-ecological
relationships, and the relationship between economic development and the environment encompasses
the relationship between poverty and the environment as well as the multidimensional aspects of
economic development. Secondly, the most popular machine learning methods are convolutional
neural networks, random forests, and support vector machines. Additionally, the most popular data
resources are satellite images, land cover, and digital elevation model data (DEM). The machine
learning methods used in each study could be classified into five types: supervised learning,
unsupervised learning, combination of unsupervised and supervised learning, combination of
unsupervised learning and natural language processing, and combination of natural language
processing and supervised learning, in which supervised learning was widely used in several themes,
especially the most notably in the environment theme. In addition, combination of supervised and
unsupervised learning was used in multiple themes. In terms of unsupervised learning, it is only solely
applied in the subcategories of agricultural landscapes and plant communities in the environment
theme, and natural language processing is only applied in the socio-economic and cultural theme.
Thirdly, according to the summary statistics, 46% of the cases studied were in Asia, while 18% and17% were in Europe and Africa, with a relatively balanced distribution. But there are relatively few cases in Central America, South America and Oceania. The
study cases were mainly concentrated in China and the United States, followed by Spain and Brazil, 86% of research published between 2018 and 2022, and the
increasing trend of the number of literatures indicates the rapid development of the field in a recent short period of time. Through the above analysis, it’s
possible to discover in depth the current progress of machine learning application in rural landscape spatial analysis, from which the research gaps and future
development trends of rural landscape intelligence can be pointed out. This study reveals the heterogeneity of data and its impact on research results, and
emphasizes the importance of standardized procedure of both data formulation and processing. At the same time, a growing need has been shown to promote
data sharing, enhance cross-regional comprehensive analysis, expand research themes and objectives, enrich data sources, and improve research methods in
future studies. With the development of machine learning technology, more innovative methods and solutions are expected to promote the sustainable
development of rural landscapes, opening up new revenue for rural landscape protection. |
| Key words: machine learning review rural landscape geospatial data landscape architecture |
