Virtual geographical scene twin modeling: a combined data-driven and knowledge-driven method with bridge construction as a case study

Jun ZhuJinbin ZhangQing ZhuLi ZuoCe LiangXiaochong ChenYakun XieFaculty of Geosciences and Engineering,Southwest Jiaotong University,Chengdu,People's Republic of China
DOI: https://doi.org/10.1080/17538947.2024.2356126
IF: 4.606
2024-05-23
International Journal of Digital Earth
Abstract:Virtual geographical scenes aim to naturally represent the real world and aid in comprehending geographic information. However, current virtual geographical scene modeling methods have problems detecting change information and low efficiency in dynamic modeling. Additionally, the modeling process lacks guidance from domain experts, resulting in poor modeling standardization. Therefore, this paper proposes a combined data-driven and knowledge-driven virtual geographical scene twin modeling method. We discuss the three-domain association geographical scene knowledge graph, change information detection network based on knowledge graph and deep learning, and knowledge-guided semantic modeling algorithms in detail. These methods enhance the ability of virtual geographical scenes to describe the changing real world. Furthermore, we select an urban bridge construction geographical scene with obvious change characteristics as a typical case, develop a prototype system, and conduct an experimental analysis. The results show that our method can fully and effectively utilize the geographical scene data and knowledge and improve the reusability of the knowledge. The average change detection accuracy in geographical scene information reaches 92.46%, and the dynamic modeling efficiency of virtual geographical twin scenes reaches 29.97 fps. Compared to other 3D modeling methods, the proposed method can represent the dynamic real world more timely, accurately, completely, and comprehensively.
geography, physical,remote sensing
What problem does this paper attempt to address?