Built Environment Reasoning from Remote Sensing Imagery Using Large Vision--Language Models

Researchers are using large language models combined with remote sensing imagery to analyze built environments for smart city applications, evaluating models like InternVL and Qwen for tasks including design suggestions, constructability assessment, and risk identification. The study demonstrates that multimodal AI systems can effectively process satellite imagery at multiple scales to support urban planning and infrastructure decision-making.

This research bridges computer vision and urban planning by exploring how advanced AI models can interpret satellite and aerial imagery to provide actionable insights about cities and infrastructure. The integration of large language models with remote sensing data represents a meaningful convergence of two powerful technologies—machine learning's reasoning capabilities with geospatial data collection—creating new possibilities for automated urban analysis.

The approach addresses a genuine need in urban development. Traditional built environment assessment requires manual inspection, expert knowledge, and significant time investment. By automating this process through multimodal AI, municipalities and developers can access rapid, scalable analysis of landuse patterns, structural feasibility, and environmental risks. The comparison between different state-of-the-art models (InternVL and Qwen) provides practical guidance on which systems deliver superior accuracy for real-world deployment.

For smart city initiatives, this capability has tangible implications. Urban planners can accelerate decision-making cycles, developers can validate project feasibility before investment, and municipalities can identify infrastructure risks more efficiently. The multi-scale imagery analysis approach is particularly valuable, as built environment reasoning requires understanding context at different resolutions—from neighborhood patterns to individual structures.

The research signals growing maturity in applying foundation models to specialized domain tasks. Success here could inspire similar applications across other infrastructure sectors including transportation networks, utility systems, and environmental monitoring. The focus on reliability and accuracy suggests the field is moving beyond proof-of-concept toward practical implementation standards that stakeholders require for mission-critical decisions.

• →Multimodal LLMs combined with remote sensing imagery can automate built environment analysis for urban planning and risk assessment.
• →InternVL and Qwen models show differential performance in accuracy and reliability for generating infrastructure recommendations.
• →Multi-scale satellite imagery analysis improves reasoning about landuse patterns and design feasibility.
• →The approach enables faster, scalable assessment of urban infrastructure compared to traditional manual evaluation methods.
• →This work demonstrates foundation models' potential for specialized domain applications beyond general-purpose language tasks.