Tong Xin-Yi, Xia Gui-Song, Zhu Xiao Xiang
Remote Sensing Technology Institute, German Aerospace Center, Münchener Straße 20, Weßling 82234, Germany.
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China.
ISPRS J Photogramm Remote Sens. 2023 Feb;196:178-196. doi: 10.1016/j.isprsjprs.2022.12.011.
High-resolution satellite images can provide abundant, detailed spatial information for land cover classification, which is particularly important for studying the complicated built environment. However, due to the complex land cover patterns, the costly training sample collections, and the severe distribution shifts of satellite imageries caused by, e.g., geographical differences or acquisition conditions, few studies have applied high-resolution images to land cover mapping in detailed categories at large scale. To fill this gap, we present a large-scale land cover dataset, . It contains more than labeled pixels of 150 high-resolution Gaofen-2 (4 m) satellite images, annotated in a 24-category system covering , , and classes. In addition, we propose a deep-learning-based unsupervised domain adaptation approach that can transfer classification models trained on labeled dataset (referred to as the ) to unlabeled data (referred to as the ) for large-scale land cover mapping. Specifically, we introduce an end-to-end Siamese network employing dynamic pseudo-label assignment and class balancing strategy to perform adaptive domain joint learning. To validate the generalizability of our dataset and the proposed approach across different sensors and different geographical regions, we carry out land cover mapping on five megacities in China and six cities in other five Asian countries severally using: PlanetScope (3 m), Gaofen-1 (8 m), and Sentinel-2 (10 m) satellite images. Over a total study area of 60,000 km, the experiments show promising results even though the input images are entirely unlabeled. The proposed approach, trained with the dataset, enables high-quality and detailed land cover mapping across the whole country of China and some other Asian countries at meter-resolution.
高分辨率卫星图像可为土地覆盖分类提供丰富、详细的空间信息，这对于研究复杂的建成环境尤为重要。然而，由于土地覆盖模式复杂、训练样本采集成本高，以及地理差异或采集条件等导致的卫星图像严重分布偏移，很少有研究将高分辨率图像应用于大规模详细类别的土地覆盖制图。为填补这一空白，我们提出了一个大规模土地覆盖数据集。它包含150幅高分二号（4米）高分辨率卫星图像的超过 个标注像素，采用24类系统进行标注，涵盖 、 和 类别。此外，我们提出了一种基于深度学习的无监督域适应方法，该方法可以将在有标注数据集（称为 ）上训练的分类模型转移到无标注数据（称为 ）上，用于大规模土地覆盖制图。具体来说，我们引入了一个端到端的连体网络，采用动态伪标签分配和类别平衡策略来进行自适应域联合学习。为了验证我们的数据集和所提出方法在不同传感器和不同地理区域的通用性，我们分别使用PlanetScope（3米）、高分一号（8米）和哨兵二号（10米）卫星图像，在中国的五个大城市和其他五个亚洲国家的六个城市进行土地覆盖制图。在总面积为60000平方公里的研究区域内，实验结果显示出良好的效果，即使输入图像完全未标注。使用 数据集训练的所提出方法，能够在米级分辨率上对中国全境和其他一些亚洲国家进行高质量、详细的土地覆盖制图。
