Sensing urban poverty: From the perspective of human perception-based greenery and open-space landscapes

Yuan Meng, Hanfa Xing, Yuan Yuan, Man Sing Wong, Kaixuan FanComput. Environ. Urban Syst 2020 · DOI

Abstract

Traditional urban poverty research often relies on top-down satellite imagery or census data, which can overlook the lived experience of residents. This study proposes a framework to sense urban poverty through "human-eye" perceptions of street-level greenery and open spaces. By applying deep learning to Baidu Street View images in Guangzhou’s Tianhe District, the researchers developed landscape indicators that correlate strongly with the Index of Multiple Deprivations (IMD). The resulting Geographically Weighted Regression (GWR) model achieved an R-squared of 0.667, demonstrating that visual landscape configurations are powerful predictors of socioeconomic status.

Introduction

Urban poverty is a multidimensional issue encompassing income, employment, education, housing, and health. While satellite data provides large-scale physical monitoring, it suffers from a "top-down" bias, failing to capture how physical barriers like walls or fences affect a resident's actual perception of their environment. This paper addresses the "knowledge gap" regarding how the street-level configuration of greenery and open space relates to urban inequality.

Methodology

The researchers developed a multi-stage framework to bridge computer vision and urban planning. The process begins with image segmentation of street-view data, followed by spatial interpolation and the calculation of specific landscape indicators.

The multi-stage framework for sensing urban poverty using street-view imagery and landscape indicators.

Image Analysis and Landscape Indicators

The Pyramid Scene Parsing Network (PSPNet) was employed to process 14,656 Baidu Street View images. The model identifies pixels belonging to "greenery" (trees, hedges) and "sky" (representing open space). From these, four key metrics were derived:

  1. Percentage of Perception: The visual field occupied by greenery or sky.
  2. Aggregation: A measure of how "clustered" or "uneven" these elements are within a neighborhood.
The PSPNet architecture used for image segmentation and the extraction of open area perceptions from street-view images.

Statistical Modeling

To account for spatial variations, the team used Geographically Weighted Regression (GWR). This allowed them to model the relationship between landscape indicators and the Index of Multiple Deprivations (IMD), which was calculated using census data across five dimensions: income, employment, education, housing, and health.

Key Insights and Results

The study revealed several critical relationships between the visual environment and urban poverty:

  • The "Wall" Effect: Satellite imagery often overestimates accessible greenery. At the street level, greenery in impoverished areas is frequently obstructed by walls, meaning residents do not perceive or benefit from it.
  • Open Space as a Poverty Indicator: Interestingly, a higher percentage of "open space" (sky) perception was positively correlated with poverty. In dense urban environments, excessive open sky at the street level often indicates a lack of complex infrastructure or "barren" landscapes rather than high-quality parks.
  • Inequality in Distribution: The "Aggregation" indicators showed that unequal, "clumpy" distributions of greenery are hallmarks of poorer neighborhoods.
  • Model Performance: The GWR model explained approximately 67% of the variation in urban poverty (R² = 0.667). Generally, visible greenery was negatively related to poverty, while visible open space was positively correlated.

Discussion and Limitations

While the study provides a robust framework for sensing poverty, it has limitations. The use of 2D images to represent 3D perception could be improved with LiDAR data. Additionally, there is a temporal lag between the 2014 census data and 2020 street-view images. Future research could also incorporate "human-in-the-loop" scoring to assess the aesthetic quality of greenery, rather than just pixel quantity.

Conclusion

By shifting the perspective from satellites to the "human eye," this research highlights the importance of the lived environment in understanding urban inequality. These findings suggest that urban planning should focus not just on the quantity of green space, but on its visibility and equitable distribution at the street level.

Summary updated

BibTeX
@article{meng2020sensing,
  author = {Yuan Meng and Hanfa Xing and Yuan Yuan and Man Sing Wong and Kaixuan Fan},
  title = {{Sensing urban poverty: From the perspective of human perception-based greenery and open-space landscapes}},
  journal = {Comput. Environ. Urban Syst.},
  year = {2020},
  doi = {10.1016/j.compenvurbsys.2020.101544},
}
Figure