Restorative perception of urban streets: Interpretation using deep learning and MGWR models

Area of Science:

• Urban Planning
• Environmental Psychology
• Geographic Information Systems (GIS)

Background:

• Restorative environments aid recovery from mental fatigue and stress.
• Urban spaces, especially streets, are crucial for city residents' well-being.
• High-density urban settings limit access to nature, increasing the importance of urban restorative functions.

Purpose of the Study:

• To develop a method for assessing the perceived restorability of urban streets using street view data and machine learning.
• To analyze the spatial heterogeneity of restorative perception in urban streets.
• To compare the effectiveness of different regression models (OLS, GWR, MGWR) in analyzing urban street restorability.

Main Methods:

• Utilized street view imagery from Shenzhen, China (Nanshan District).
• Developed a deep learning scoring model and employed SegNet for visual element classification.
• Applied a random forest algorithm and multiscale geographically weighted regression (MGWR) for restorative perception analysis.

Main Results:

• Spatial heterogeneity was observed in the restorative perception of urban streets.
• The MGWR model demonstrated higher explanatory power (R²) compared to OLS and GWR.
• MGWR allowed for detailed analysis of individual visual street elements' impact on restorability.

Conclusions:

• The developed deep learning and MGWR approach effectively assesses urban street restorability.
• Findings highlight the importance of urban street design in promoting psychological well-being.
• Results can inform design guidelines for creating more restorative urban environments.