Semantic Digital Twin Modeling of Urban Elements: Methods and Applications in Smart Cities

ABSTRACT Digital twin technologies are reshaping smart city development by enabling precise and efficient urban modeling. However, traditional photogrammetry‐based 3D reconstruction methods face challenges such as high computational costs, redundant data, and limited semantic details, which restrict their scalability. To overcome these limitations, we propose an automated framework for modeling urban elements, including buildings, water bodies, roads, and vegetation, using oblique photogrammetry data. The framework integrates advanced AI‐driven semantic segmentation, geometric simplification, horizontal slicing for building extraction, and texture mapping to produce lightweight, individualized models. A key innovation of this framework lies in its precise modeling of both vegetation and buildings. For vegetation, the framework employs neural network classification, random point distribution, and 3D engine‐based elevation sampling to generate accurate digital twin models. For buildings, horizontal slicing is used to efficiently reconstruct 3D structures by segmenting and refining characteristic surfaces. This approach reduces data volume by over 80% while preserving visual fidelity, facilitating large‐scale scene rendering and interactive simulations. The proposed framework has been successfully applied to the Jingdezhen digital twin project, covering 1300 km 2 , demonstrating its effectiveness in large‐scale urban reconstruction. Applications include urban planning, ecological monitoring, and low‐altitude economic activities such as drone logistics. By enhancing automation, scalability, and semantic integration, this framework provides a robust foundation for building digital twin cities to support urban management and decision‐making.