Page 160 - 《武汉大学学报（信息科学版）》2025年第9期

第 50 卷 第 9 期 武 汉 大 学 学 报（ 信 息 科 学 版 ） Vol.50 No.9
2025 年 9 月 Geomatics and Information Science of Wuhan University Sept. 2025

引文格式：刘梓航,刘欣怡,张永军 . 联合全局-局部特征和动态错误监督的遥感影像建筑物损伤评估方法[J]. 武汉大学学报
（信息科学版）,2025,50(9):1888-1903.DOI:10.13203/j.whugis20240346
Citation：LIU Zihang,LIU Xinyi,ZHANG Yongjun.Building Damage Assessment from Satellite Images Combining Global-Lo⁃
cal Features and Dynamic Error Supervision[J].Geomatics and Information Science of Wuhan University,2025,50(9):1888-1903.
DOI:10.13203/j.whugis20240346

联合全局-局部特征和动态错误监督的

遥感影像建筑物损伤评估方法

刘梓航 刘欣怡 张永军 1
1
1
1 武汉大学遥感信息工程学院，湖北 武汉，430079
摘 要：在灾害发生后，快速准确地评估灾害区域的范围和严重程度对于后续的救援和重建至关重要。目前针对遥感影
像的深度学习建筑物损伤评估方法面临特征差异建模不足、全局-局部特征利用不充分和困难样本感知能力缺乏等问
题。为此，提出一种基于全局-局部特征融合和动态错误监督网络（global-local feature fusion and dynamic error supervi⁃
sion network，GLESNet）的双时相遥感影像建筑物损伤评估方法。在编码阶段，采用共享权重的编码器提取双时相影像
特征，将双时相影像特征送入差异增强融合模块，增强特征间差异并获取融合特征；在解码阶段，融合特征先后经过全
局-局部特征融合模块和动态错误感知解码器并输出评估结果，以实现兼顾全局-局部特征的融合解码和困难样本感知
的学习。在目前最大的全球级别建筑物损伤评估高分辨率遥感影像数据集 xBD 上进行实验，GLESNet 取得了建筑提取
F1 分数 86.03%，损伤分类 F1 分数 75.20%，综合评价 F1 分数 78.45% 的结果，总体指标优于多个对比方法。在 Ida-BD
和 LEVIR-CD 数据集上进行了迁移实验和变化检测实验，验证了 GLESNet 的泛化性和不同任务适用性。
关键词：建筑物损伤评估；变化检测；孪生网络；特征融合；遥感影像
中图分类号：P237 文献标识码：A 收稿日期：2024⁃11⁃18
DOI：10.13203/j.whugis20240346 文章编号：1671⁃8860（2025）09⁃1888⁃16

Building Damage Assessment from Satellite Images Combining Global-
Local Features and Dynamic Error Supervision

LIU Zihang LIU Xinyi ZHANG Yongjun 1
1
1
1 School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China
Abstract： Objectives: After a disaster, it is essential to quickly and accurately assess the extent and severi⁃
ty of disaster area for subsequent humanitarian relief and reconstruction. Traditional damage assessment
methods are constrained by time efficiency, labor cost, and accessibility. In contrast, satellite images can
quickly obtain the real situation of a wide range of disaster area, and gradually become an important data
source for building damage assessment. Automated building damage assessment from satellite images relies
on deep learning methods, but current deep learning methods face challenges such as insufficient modeling
of feature differences, inadequate utilization of global-local features, and lack of difficult sample perception
ability. Methods: To address these problems, a building damage assessment method based on global-local
feature fusion and dynamic error supervision network (GLESNet) is proposed. At the encoding stage, the
dual-temporal image features were extracted by a shared weight backbone, and the features were sent to
the difference enhancement fusion module to enhance the difference between the features, filter out spuri⁃
ous changes, and obtain the fusion features. At the decoding stage, the fusion features are passed by the
vertical and horizontal global-local feature fusion modules and the dynamic error aware decoder, to fuse the

基金项目：国家自然科学基金（42192581，42201474）。
第一作者：刘梓航，硕士生，研究方向为地表异常遥感检测与评估。zhliu2022@whu.edu.cn
通信作者：刘欣怡，博士，副研究员。liuxy0319@whu.edu.cn