中文 | English

J. Hefei Univ. Tech. (Nat. Sci.)

合肥工业大学学报(自然科学版)

JOURNAL OF HEFEI UNIVERSITY OF TECHNOLOGY (NATURAL SCIENCE)

Competent Authorities: Ministry of Education of the People's Republic of China

Sponsor by: Hefei University of Technology

ISSN 1000-1646, CN 34-1001/N

High-resolution remote sensing image building extraction based on improved Swin-UNet network

Download Full Text

基于改进 Swin-UNet 网络的高分辨率遥感影像建筑物提取

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Nov.2024, Vol.47 No.11: 1571-1579

DOI: 10.3969/j.issn.1003-5060.2024.11.021

Authors

YUAN Xiaoyu, LI Zhenxuan, GAO Fei

(School of Civil and Hydraulic Engineering, Hefei University of Technology, Hefei 230009, China)

Abstract and Keywords

Abstract: In view of the low automation level of traditional artificial extraction methods in building extraction tasks, and the poor and incomplete edge extraction of existing deep learning methods such as convolutional neural network (CNN) and UNet in remote sensing image building extraction, this paper presents an automatic building extraction method based on improved Swin-UNet network model. Based on the original Swin-UNet network structure, the new network model uses cross-attention block (CAB) to replace the Swin Transformer block of the original network to build a new network system. The applicability of the model is verified in the building extraction experiments of Wuhan University (WHU) Aerial Building Dataset and Massachusetts Buildings Dataset. The results show that the proposed method is superior to support vector machine (SVM) algorithm and traditional deep learning methods, and has good segmentation accuracy and robust generalization ability.

Keywords: high-resolution remote sensing image; deep learning; building extraction; Swin-UNet network structure

作者信息

袁啸宇，李振轩，高飞

（合肥工业大学土木与水利工程学院，安徽合肥230009）

摘要和关键词

摘要: 针对传统人工提取方法目前在建筑物提取任务中自动化水平低,以及现有的卷积神经网络(convolutional neural network, CNN)、UNet等深度学习方法在遥感影像建筑物提取中边缘提取效果差、提取不完整等问题,文章提出一种基于改进的Swin-UNet网络模型的建筑自动提取方法。新网络模型在原Swin-UNet网络结构基础上,采用跨块注意力机制(cross-attention block, CAB)取代原网络的Swin Transformer块来构建新的网络体系,在武汉大学航空(WHU)建筑数据集和美国马萨诸塞州建筑物数据集建筑物提取试验中验证了模型的适用性。研究结果表明,该方法优于支持向量机(support vector machine, SVM)算法及基于传统的深度学习方法,具有良好的分割精度和鲁棒泛化能力。

关键词: 高分辨率遥感影像；深度学习；建筑物提取；Swin-UNet网络结构

Funding

安徽省自然科学基金资助项目(2208085QD105);中央高校基本科研业务费专项资金资助项目(JZ2021HGTA0167)

Previous Next

Back to Contents