基于U2-Net的遥感影像水体提取及其在雨洪韧性评估中的应用  

Remote sensing image water body extraction based on the U2-Net model and its application in flood resilience assessment

作　　者：周敏 韩邹禹 杨紫曦 胡景瑞 黄明华[2] 陈力 ZHOU Min;HAN Zouyu;YANG Zixi;HU Jingrui;HUANG Minghua;CHEN Li(Shaanxi Xixian Urban and Rural Development Group Co.,Ltd.,Xi'an 712000,China;School of Building Equipment Science and Engineering,Xi’an University of Architecture and Technology,Xi'an 710055,China)

机构地区：[1]陕西西咸城乡发展集团有限公司,陕西西安712000 [2]西安建筑科技大学建筑设备科学与工程学院,陕西西安710055

出　　处：《经纬天地》2025年第2期58-63,共6页Survey World

基　　金：国家自然科学基金青年基金(52208126);西安建筑科技大学交叉研究培育专项(1960523134/2023)支持。

摘　　要：针对全球气候变化加剧极端天气、雨洪灾害频发的现状,提出了一种图像语义分割模型的遥感影像水体提取方法,旨在提升雨洪韧性评估的时效性和准确性。采用具有多尺度级联机制的U2-Net模型,利用Landsat 8遥感数据,建立了一个包含预处理、模型训练和结果分析的综合性水体提取流程,并通过与传统方法的对比,验证了U2-Net模型在水体提取中的优势。在不同季节的图像中均能精确识别水体边界,即使在复杂水体形态下也能保持高识别稳定性,体现了模型的泛化能力和鲁棒性。此外,模型的F_(1)-Score达到了0.902,MAE为0.008,处理单幅图像的平均时间为0.058 s,表明其在实时遥感提取中的高效潜力。综上,通过U2-Net模型的应用,不仅提高了水体提取的精度和效率,还为雨洪韧性评估及水资源管理提供了有力的数据支持,展现了深度学习在环境监测领域的广泛应用前景。In response to the worsening global climate change,increasing extreme weather events and frequent rainstorm-induced floods,a remote sensing image water body extraction method is proposed in this paper by using an image semantic segmentation model to aim at enhancing the timeliness and accuracy of flood resilience assessment.In this study,the U2-Net model is employed,which features a multi-scale cascading mechanism,utilizing Landsat 8 remote sensing data to establish a comprehensive water body extracting process including preprocessing,model training and result analysis.Compared with the traditional method,the advantages of U2-Net model in water body extraction get verified.The model can accurately recognize the water boundaries in images from different seasons and maintain high recognition stability even under complex water body morphologies,demonstrating the generalization ability and robustness of the model.Additionally,an F_(1)-Score of the model reached 0.902,a Mean Absolute Error(MAE)is 0.008,and an average processing time for each image is 0.058 seconds,highlighting its potential for real-time remote sensing monitoring.In summary,the application of the U2-Net model not only improves the accuracy and efficiency of water body extraction but also provides robust data support for flood resilience assessment and water resource management.Meanwhile it showcases the broad application prospects of deep learning in the field of environmental monitoring.

关 键 词：卫星遥感影像 水体提取 语义分割 机器学习 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]