KU-Net:改进U-Net的高分辨率遥感影像建筑物提取方法  

KU-Net:An Improved U-Net Method for Building Extraction from High Resolution Remote Sensing Imagery

作　　者：刘卓涛 龚循强[1,2,3] 夏元平 陈晓勇 吴晋涛[1,2] LIU Zhuotao;GONG Xunqiang;XIA Yuanping;CHEN Xiaoyong;WU Jintao(Key Laboratory of Mine Environmental Monitoring and Improving Around Poyang Lake of Ministry of Natural Resources,East China University of Technology,Nanchang 330013,China;School of Surveying and Geoinformation Engineering,East China University of Technology,Nanchang 330013,China;State Key Laboratory of Information Engineering in Surveying,Mapping,and Remote Sensing,Wuhan University,Wuhan 430079,China)

机构地区：[1]东华理工大学自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室,南昌330013 [2]东华理工大学测绘与空间信息工程学院,南昌330013 [3]武汉大学测绘遥感信息工程国家重点实验室,武汉430079

出　　处：《遥感信息》2024年第5期121-131,共11页Remote Sensing Information

基　　金：国家自然科学基金(42101457、42174055);自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室开放基金重点项目(MEMI-2023-01);武汉大学测绘遥感信息工程国家重点实验室开放研究基金(23R03)。

摘　　要：针对遥感影像背景复杂导致的建筑物提取结果存在边界模糊、小目标漏检、地物误检等问题,提出一种基于改进U-Net的建筑物提取网络KU-Net(Keep border U-Net)。该网络在U-Net的基础上加入空洞空间金字塔池化和含注意力机制的横向连接模块,其中,空间空洞金字塔池化能够提升模型的感受野,横向连接模块对不同层级的跳跃连接特征进行融合,缓解特征丢失的情况,从而进一步提高精度。实验结果表明,该方法相比于其他对比方法,提取结果更为清晰准确,对边缘有较好的保持效果,定量结果更优。In order to solve these problems of boundary blurring,missed detection of small targets,and false detection of ground objects caused by complex background in remote sensing images,a building extraction network,KU-Net(keep border U-Net),is proposed in this paper.The network incorporates Atrous spatial pyramid pooling and a lateral connection module with attention mechanism on the basis of U-Net.The Atrous spatial pyramid pooling can enhance the receptive field of the model,and the horizontal connection module integrates skip connection features at different levels to alleviate the problem of feature loss and further improve accuracy.The experimental results show that the proposed method in this paper has clearer and more accurate extraction results,better edge preservation effect and better quantitative results compared with other comparative methods.

关 键 词：KU-Net 建筑物提取 空洞空间金字塔池化 WHU建筑物数据集 注意力机制 

分 类 号：P237[天文地球—摄影测量与遥感]