云南冶金2025，Vol.54Issue(6)：8-15,8.
基于SegFormer语义分割模型的复杂地区建筑物提取方法研究
Research on Building Extraction Methods in Complex Areas Based on the SegFormer Semantic Segmentation Model
摘要
Abstract
To address the issues of insufficient recognition accuracy and blurred boundaries in remote sensing images of plateau regions,caused by complex terrain,intense illumination,and sparse building distribution,this study proposes an improved SegFormer semantic segmentation model based on remote sensing images of Yuanyang County,Yunnan Province,and data from the Third National Land Survey.The model integrates an Enhanced Spatial Feature Fusion Module(SFFM)and a Context-Aware Gated Attention Module(CGAM).The research results indicate that the improved model achieves an mIoU of 84.34％and an mPA of 94.5％,significantly outperforming mainstream models such as U-Net and DeepLabV3+in terms of performance.The method demonstrates stronger robustness in small object detection and complex boundary restoration,making it well-suited for building extraction tasks in complex plateau environments and holding promising application prospects.关键词
SegFormer/语义分割/建筑物提取/遥感影像/高原地区Key words
SegFormer/semantic segmentation/building extraction/remote sensing imagery/plateau region分类
信息技术与安全科学引用本文复制引用
付小东,朱有建,赵福超,李斯曼,郑乾璐,吕双铜..基于SegFormer语义分割模型的复杂地区建筑物提取方法研究[J].云南冶金,2025,54(6):8-15,8.基金项目
云南省有色地质局三○八队科创基金项目"基于遥感的生态保护红线保护成效评估技术方法研究"(2024-30804) （2024-30804）
