Protocol for assessing neighborhood physical disorder using the YOLOv8 deep learning model

Area of Science:

• Urban planning and public health
• Environmental science
• Computer vision and artificial intelligence

Background:

• Neighborhood physical disorder (PD) is linked to adverse economic, social, and public health outcomes.
• Existing methods for assessing PD lack quantitative standardization and scalability.
• Understanding and quantifying PD is crucial for targeted urban interventions.

Purpose of the Study:

• To present a standardized protocol for the quantitative assessment of neighborhood physical disorder (PD).
• To establish a reproducible methodology for measuring PD using deep learning and street-level imagery.
• To provide a foundation for cross-regional and international comparisons of PD.

Main Methods:

• Collecting street view imagery across diverse urban environments.
• Developing and implementing object detection models, specifically YOLOv8, for identifying PD indicators.
• Calculating quantitative PD scores based on detected elements and spatial metrics.
• Analyzing changes in PD over time and across different geographical scales.

Main Results:

• A validated protocol for quantitative PD assessment was successfully developed.
• The methodology allows for objective and reproducible measurement of PD indicators.
• The approach demonstrated the ability to quantify PD variations across streets and cities.

Conclusions:

• The proposed protocol offers a robust and scalable method for assessing neighborhood physical disorder.
• This quantitative approach can inform urban planning, public health strategies, and environmental justice initiatives.
• The methodology provides a foundation for future research on the impacts of PD in diverse global contexts.