Smart cities get smarter: AI-powered material detection for sustainable urban planning

The construction sector is a major driver of global carbon emissions, with buildings alone responsible for nearly one-third of worldwide energy-related CO₂ emissions. However, existing methods for assessing building materials are often constrained by limited geographic scope, poor scalability, and insufficient accuracy. Conventional databases struggle to provide comprehensive material intensity assessments, especially across diverse urban landscapes. These challenges underscore the urgent need for innovative, data-driven solutions that can deliver precise and actionable insights at scale.

A collaborative research initiative led by Peking University and the University of Southern Denmark has risen to this challenge. The team has developed an advanced framework that integrates deep learning with remote sensing to identify building materials with unprecedented precision. Their findings (DOI: 10.1016/j.ese.2025.100538), published on February 3, 2025, in Environmental Science and Ecotechnology, showcase the potential of this technology in creating customized material intensity databases tailored to different urban regions, paving the way for more sustainable and efficient city planning.

The study employs a fusion of Google Street View imagery, satellite data, and OpenStreetMap geospatial information to classify building materials with high accuracy. By leveraging Convolutional Neural Networks (CNNs), the researchers trained models capable of identifying roof and façade materials with exceptional detail. The models were first trained using extensive datasets from Odense, Denmark, before being successfully validated in major Danish cities such as Copenhagen, Aarhus, and Aalborg. The validation process confirmed the framework’s robustness, demonstrating its ability to generalize across diverse urban settings and reinforcing its scalability.

A key innovation of the study is its use of advanced visualization techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM), which offers a window into how the AI models interpret imagery. By revealing which parts of an image most influence classification decisions, this technique enhances model transparency and reliability. Additionally, the researchers developed material intensity coefficients to quantify the environmental impact of different building materials. By combining high-resolution imagery with deep learning, this framework overcomes longstanding limitations in material data availability and accuracy, providing a powerful tool for sustainable urban development.

Highlights

• A scalable framework supports the creation of customized material intensity databases for diverse regions, facilitating sustainable urban planning and retrofits.

• Deep learning enables precise identification of building materials using remote sensing and street view data.

• Visualizations of model predictions enhance interpretability and reveal decision-making processes.

• Accurate material assessments inform targeted building upgrades for improved energy efficiency.

Prof. Gang Liu, the principal investigator of this project, highlighted the transformative potential of the technology: “Our study demonstrates how deep learning and remote sensing can fundamentally change the way we analyze and manage urban building materials. With precise material intensity data, we can drive more sustainable urban planning and targeted retrofitting, contributing directly to global carbon reduction efforts.”

The implications of this breakthrough extend far beyond academic research. By enabling cities to accurately identify and map building materials, this framework equips urban planners with critical data for energy efficiency strategies, carbon reduction policies, and circular economy initiatives. Its scalability ensures that the approach can be adapted to different urban environments, making it a game-changer for sustainable city planning worldwide.

###

References

DOI

10.1016/j.ese.2025.100538

Original Source URL

https://doi.org/10.1016/j.ese.2025.100538

Funding information

This work is financially supported by the National Natural Science Foundation of China (71991484, 71991480), the Fundamental Research Funds for the Central Universities of Peking University, the Independent Research Fund Denmark (iBuildGreen), the European Union under grant agreement No. 101056810 (CircEUlar), and the China Scholarship Council (202006730004 and 202107940001).

About Environmental Science and Ecotechnology

Environmental Science and Ecotechnology (ISSN 2666-4984) is an international, peer-reviewed, and open-access journal published by Elsevier. The journal publishes significant views and research across the full spectrum of ecology and environmental sciences, such as climate change, sustainability, biodiversity conservation, environment & health, green catalysis/processing for pollution control, and AI-driven environmental engineering. The latest impact factor of ESE is 14, according to the Journal Citation ReportTM 2024.