Related Experiment Video
Updated: Apr 23, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
11.0K
Image-based prediction of residential building attributes with deep learning
Weimin Huang1, Alexander W Olson2, Elias B Khalil1
1Department of Mechanical and Industrial Engineering, University of Toronto, Toronto, ON Canada.
Journal of Industrial Ecology
|April 22, 2026
Summary
Machine learning accurately estimates building floor area and age from street-view images. This advances urban metabolism studies by providing crucial data for material flow and greenhouse gas analysis.
Area of Science:
- Urban planning and remote sensing
- Environmental science and sustainability
- Computer vision and machine learning
Background:
- Building age and floor area are critical for understanding urban metabolism, material flows, and embodied greenhouse gas (GHG) emissions.
- Traditional methods for collecting this data are often unreliable, uneven, and costly.
- Accurate building data is essential for effective resource management and environmental impact assessment in urban areas.
Purpose of the Study:
- To develop and evaluate an image-based machine learning approach for estimating building floor area and age.
- To address the limitations of traditional survey methods in data acquisition for urban studies.
- To provide a scalable solution for generating building attribute data for large-scale analyses.
Main Methods:
- Building attributes (floor area and age) were estimated using machine learning models trained on Google Street View images.
- Area prediction was formulated as a regression problem, and age prediction as a classification problem across six historical periods.
- An EfficientNetV2 module was employed for feature extraction, followed by fully connected layers for attribute estimation.
Main Results:
- The model achieved a mean absolute percentage error of 19.42% for floor area prediction and 70.27% accuracy for age prediction in Toronto.
- Performance varied across five other Canadian cities, demonstrating the importance of local training data.
- The study confirmed the feasibility of using street-view imagery for automated building attribute estimation.
Conclusions:
- Machine learning offers a viable and automated method for estimating key building attributes from readily available street-view imagery.
- This approach can significantly improve the availability and accuracy of data for urban metabolism, material flow, and embodied GHG studies.
- The findings support the development of large-scale, data-driven analyses for sustainable urban development and environmental management.