机电工程技术2025，Vol.54Issue(6)：28-33,39,7.DOI:10.3969/j.issn.1009-9492.2025.00014
基于高效局部注意力和全局上下文增强的遥感影像建筑物提取方法
A Building Extraction Method for Remote Sensing Images Based on Efficient Local Attention and Global Context Enhancement
摘要
Abstract
As the main component of the city,the study of accurate and efficient extraction of building information from high-resolution remote sensing images is of great significance for urban planning,land use,and disaster assessment.However,when semantic segmentation means are used for building information extraction,problems such as low segmentation accuracy,omission,and wrong extraction are often present.In addressing the issues above,a global context-enhanced feature extraction network is proposed based on the U-Net network to enhance the extraction of building context-detail features in the feature extraction phase and reduce the possibility of false detection;meanwhile,the efficient local attention is introduced to achieve the accurate differentiation of the building regions and the problem of fine extraction of buildings under complex background is solved.To verify the effectiveness of the method,experiments are conducted on two building datasets and compared with mainstream semantic segmentation methods.In the experimental section of the WHU dataset,the proposed methodology attained IoU,Precision,Recall,and F1 scores of 90.21%,94.96%,94.74%,and 94.85%,respectively.It is significantly higher than the other comparative networks and the effect of building segmentation in the resultant graphs is more refined,and the universality of the network is also verified in the Guiyang dataset.关键词
遥感影像/建筑物提取/深度学习/Unet/注意力机制Key words
remote sensing imagery/building extraction/deep learning/Unet/attention mechanism分类
测绘与仪器引用本文复制引用
张永城,刘春阳,刘裕芸,王德金..基于高效局部注意力和全局上下文增强的遥感影像建筑物提取方法[J].机电工程技术,2025,54(6):28-33,39,7.基金项目
矿山采动灾害空天地协同监测与预警安徽普通高校重点实验室(安徽理工大学)开放基金资助(KLAHEI202203) （安徽理工大学）
