Image-based machine learning and cluster analysis for urban road network: employing Orange for codeless visual programming
Zuo Zhang,Mengwei Zhang,Xiangxiang Song,Zhi Li
DOI: https://doi.org/10.1080/10095020.2024.2377212
IF: 4.278
2024-07-19
Geo-spatial Information Science
Abstract:In the pursuit of sustainable urban development, it is imperative to address strategies for optimizing and harmonizing the configuration of urban spatial layouts, especially in response to rapid urbanization. Urban road networks play a pivotal role in shaping intricate urban spatial patterns as they intricately weave throughout the cityscape, significantly influencing land utilization, spatial arrangements, and the mobility of residents. The urban environment is a complex system teeming with multifaceted human activities. With the continuous evolution of advanced data and computational technologies, it is now feasible to construct intricate models for an in-depth exploration of urban spatial structures. In this study, we constructed an analytical framework that leveraged road network image recognition, multi-scale influence factors analysis, and predictions using codeless visual programming. Within this framework, deep learning and unsupervised clustering are applied to complete the hierarchical clustering of road networks and exploit the intrinsic association between cities. Meanwhile, hierarchical clustering and natural and humanistic attributes of cities are connected to investigate the strength of different attributes on specific clusters. The research has the following results: (1) In addition to the alignment of urban road network clustering outcomes, quantitative unevenness and geographical heterogeneity are also observed. (2) Multiple attributes have a complicated impact on urban space, including city size, and socioeconomic and geographic characteristics. Among these factors, socio-economic variables emerge as the principal drivers influencing urban spatial structure. (3) The research enhances the model's generalization capability and predictive accuracy. As attributes increase and model learning improves, both the Area Under Curve (AUC) and F-1 scores exhibit continuous enhancements, ultimately achieving error-free predictions. The study's findings offer valuable insights into urban spatial structures and systems, supporting urban decision-makers in addressing challenges associated with urban growth.
remote sensing