当前位置: X-MOL 学术Int. J. Appl. Earth Obs. Geoinf. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
DSA-Net: A novel deeply supervised attention-guided network for building change detection in high-resolution remote sensing images
International Journal of Applied Earth Observation and Geoinformation ( IF 7.6 ) Pub Date : 2021-10-22 , DOI: 10.1016/j.jag.2021.102591
Qing Ding 1 , Zhenfeng Shao 1 , Xiao Huang 2 , Orhan Altan 3
Affiliation  

Building change detection (BCD) plays a crucial role in urban planning and development and has received extensive attention. However, existing deep learning-based change detection methods suffer from limited accuracy, mainly due to the information loss and inadequate capability in feature extraction. To overcome these shortcomings, we propose a novel deeply supervised attention-guided network (DSA-Net) for BCD tasks in high-resolution images. In the DSA-Net, we innovatively introduce a spatial attention mechanism-guided cross-layer addition and skip-connection (CLA-Con-SAM) module to aggregate multi-level contextual information, weaken the heterogeneity between raw image features and difference features, and direct the network’s attention to changed regions. We also introduce an atrous spatial pyramid pooling (ASPP) module to extract multi-scale features. To further improve detection performance, we implement a new deep supervision module to enhance the ability of middle layers to extract more distinctive features. We conduct quantitative and qualitative experiments on the two publicly available datasets, i.e., the LEVIR-CD and the WHU Building datasets. Compared with the competing methods, the proposed DSA-Net achieves the best performance in all evaluation metrics. The efficiency analysis reveals that the proposed DSA-Net achieves a great balance between BCD performance and complexity/efficiency, with faster convergence and higher robustness.



中文翻译:

DSA-Net:一种新颖的深度监督注意力引导网络,用于在高分辨率遥感图像中构建变化检测

建筑变化检测(BCD)在城市规划和发展中起着至关重要的作用,受到了广泛的关注。然而,现有的基于深度学习的变化检测方法精度有限,主要是由于信息丢失和特征提取能力不足。为了克服这些缺点,我们为高分辨率图像中的 BCD 任务提出了一种新颖的深度监督注意力引导网络(DSA-Net)。在DSA-Net中,我们创新地引入了空间注意力机制引导的跨层加法和跳跃连接(CLA-Con-SAM)模块来聚合多级上下文信息,弱化原始图像特征和差异特征之间的异质性,并将网络的注意力引向变化的区域。我们还引入了一个多孔空间金字塔池化 (ASPP) 模块来提取多尺度特征。为了进一步提高检测性能,我们实现了一个新的深度监督模块,以增强中间层提取更鲜明特征的能力。我们对两个公开可用的数据集进行了定量和定性实验,即 LEVIR-CD 和 WHU Building 数据集。与竞争方法相比,所提出的 DSA-Net 在所有评估指标中都取得了最佳性能。效率分析表明,所提出的 DSA-Net 在 BCD 性能和复杂性/效率之间取得了很好的平衡,收敛速度更快,鲁棒性更高。我们实现了一个新的深度监督模块,以增强中间层提取更多鲜明特征的能力。我们对两个公开可用的数据集进行了定量和定性实验,即 LEVIR-CD 和 WHU Building 数据集。与竞争方法相比,所提出的 DSA-Net 在所有评估指标中都取得了最佳性能。效率分析表明,所提出的 DSA-Net 在 BCD 性能和复杂性/效率之间取得了很好的平衡,收敛速度更快,鲁棒性更高。我们实现了一个新的深度监督模块,以增强中间层提取更多鲜明特征的能力。我们对两个公开可用的数据集进行了定量和定性实验,即 LEVIR-CD 和 WHU Building 数据集。与竞争方法相比,所提出的 DSA-Net 在所有评估指标中都取得了最佳性能。效率分析表明,所提出的 DSA-Net 在 BCD 性能和复杂性/效率之间取得了很好的平衡,收敛速度更快,鲁棒性更高。

更新日期:2021-10-22
down
wechat
bug