测绘科学技术学报2025，Vol.41Issue(6)：619-625,7.DOI:10.3969/j.issn.1673-6338.2025.06.010
结合U-Net和分割基础模型的快速建筑物自动提取方法
A Fast Automatic Building Extraction Method Combining U-Net and Fundamental Segmentation Model
摘要
Abstract
The automatic extraction of buildings from remote sensing imagery is of great significance for tasks such as population estimation,urban planning,and sociological analysis.However,most of the existing deep learning methods for semantic segmentation of buildings have problems such as limited precision and smooth boundaries of extracted buildings.To address these challenges,a fast automatic building extraction method that integrates U-Net with a fundamental segmentation model is proposed,leveraging the advantages of both frameworks.Firstly,the coarse segmentation results are generated by U-Net.Subsequently,the point prompts are extracted from the result and input into the pre-trained fundamental segmentation model EfficientViT-SAM to iteratively refine the semantic segmentation results of U-Net.The accuracy of the method is verified using the WHU building dataset.Compared with U-Net,the intersection-over-union ratio,precision,and recall rates are enhanced from 68.42％,83.11％,and 89.11％to 78.92％,91.20％,and 89.73％,respectively.In contrast to the method combining U-Net and the classical ViT-H pre-trained model,the proposed method in this paper significantly improves the running speed while achieving the same level of accuracy.关键词
深度学习/U-Net模型/迭代式策略/分割基础模型/建筑物提取Key words
deep learning/U-Net/iterative strategy/fundamental segmentation model/building extraction分类
天文与地球科学引用本文复制引用
莫加伟,万江琴,王凯华..结合U-Net和分割基础模型的快速建筑物自动提取方法[J].测绘科学技术学报,2025,41(6):619-625,7.基金项目
自然资源部中国地质调查局项目(DD20230515 （）
DD20220874). （）
