基于改进 SwinU-Net 遥感影像洪涝灾害信息提取与变化分析

利用遥感技术对影像进行信息提取是检测地表信息变化的一种重要手段, 在城市发生洪涝灾害时的监测及灾后评估方面发挥着重要作用。在高分辨率遥感影像地表信息提取过程中, 针对基础 SwinU-Net 网络精度不够高且出现地物边缘提取不清晰、小型地物提取不完整等问题, 文章提出一种基于改进 SwinU-Net 网络的地表信息提取方法, 并在 2021 年 7 月河南暴雨洪涝灾害遥感影像信息提取实验中得到验证。结果表明, 该方法优于基础 SwinU-Net 提取方法及其他深度学习方法, 能进一步提升分类精度, 可有效完成洪涝灾害信息提取。

The extraction of information from images using remote sensing technology is an important means of detecting changes in surface information, and it plays a significant role in monitoring and post-disaster assessment of urban flood disasters. In the process of extracting surface information from high-resolution remote sensing images, there are issues with the baseline SwinU-Net network such as insufficient accuracy, unclear extraction of object edges, and incomplete extraction of small objects. To address these problems, a surface information extraction method based on an improved SwinU-Net network is proposed, and it was validated through experiments on remote sensing image information extraction during the heavy rain and flood disaster in Henan Province, China in July 2021. The experimental results demonstrate that this method outperforms the baseline SwinU-Net extraction method as well as several other deep learning methods, thereby improving the classification accuracy and effectively achieving flood disaster information extraction.