Building Extraction Model Based on UAV Remote Sensing Image: A Case Study of Bohai Bay Area in Shandong Province, China
Open Access
- 1 January 2026
- journal article
- Published by Wiley in Advances in Civil Engineering
- Vol. 2026 (1) , 3116626
Abstract
To address the issues in traditional earthquake disaster assessment, such as low efficiency, high risk of manual surveys, and limitations of two‐dimensional data, this study selects 16 districts and counties in the Bohai Bay area of Shandong Province (with a total area of 23,635 square kilometers, over 90% of which are rural areas) as the research area. Based on 32,854 high‐resolution images (with a resolution of 0.02–0.046 m) obtained by drones from 2023 to 2024 and on‐site survey data of 8153 buildings, a Mask R‐CNN deep learning model combined with the nDSM is constructed. The data are preprocessed through cloth simulation filter (CSF) filtering and the visible band difference vegetation index (VDVI) to remove vegetation and ground interference, enabling the automatic extraction of building outlines, structural types, heights, and building areas. Rural residences in the research area are mainly of brick‐wood structure (59%) and brick‐concrete structure, with significant regional structural differences. Experiments show that the overall recognition accuracy of the model for building structural types reaches 92.69%, among which the recognition accuracy for brick‐wood structures is 92.05%. The recognition accuracy for 208 buildings is 100%, and the average error rate of building area extraction is only 0.063. This method integrates ArcGIS spatial analysis and Python tools, overcomes the limitations of traditional methods, provides efficient and accurate technical support for preassessment of earthquake losses, seismic reinforcement of rural buildings, and disaster management, and also offers a reference for building information extraction in similar earthquake‐prone areas.Keywords
This publication has 12 references indexed in Scilit:
- Rapid Assessment of Earthquake-Induced Building Damage Using Optical Image CorrelationIEEE Transactions on Geoscience and Remote Sensing, 2026
- UEBNet: A Novel and Compact Instance Segmentation Network for Post-Earthquake Building Assessment Using UAV ImageryRemote Sensing, 2025
- Characterizing and classifying urban tree species using bi-monthly terrestrial hyperspectral images in Hong KongISPRS Journal of Photogrammetry and Remote Sensing, 2021
- Application of Deep Learning and Unmanned Aerial Vehicle on Building MaintenanceAdvances in Civil Engineering, 2021
- UAV and GIS based rapid earthquake-induced building damage assessment and methodology for EMS-98 isoseismal map drawing: The June 12, 2017 Mw 6.3 Lesvos (Northeastern Aegean, Greece) earthquakeInternational Journal of Disaster Risk Reduction, 2019
- Feature Pyramid Networks for Object Detection2017
- Reviews of unmanned aerial vehicle (drone) technology trends and its applications in the mining industryGeosystem Engineering, 2016
- Fully convolutional networks for semantic segmentation2015
- Earthquake damage detection using remote sensing data2007
- Global land cover classifications at 8 km spatial resolution: The use of training data derived from Landsat imagery in decision tree classifiersInternational Journal of Remote Sensing, 1998