MLRSNet

MLRSNet是由中国地质大学(北京)和北京师范大学等机构合作创建的高分辨率多标签遥感数据集，旨在通过深度学习技术提升遥感图像的语义场景理解能力。该数据集包含109,161张来自全球各地的高分辨率光学卫星或航空图像，涵盖46个场景类别，每张图像至少关联60个预定义标签中的一个至十三个。数据集的构建过程包括场景样本收集、数据库质量控制和样本多样性改进，确保了数据的高质量和广泛代表性。MLRSNet特别适用于多标签图像分类和检索任务，通过与现有数据集如ImageNet的结合，能有效解决对象旋转、类内变异和类间相似性等挑战，为遥感图像分析提供了强大的基准。

MLRSNet is a high-resolution multi-label remote sensing dataset jointly developed by China University of Geosciences (Beijing), Beijing Normal University and other institutions, aiming to enhance the semantic scene understanding capability of remote sensing images through deep learning technologies. This dataset contains 109,161 high-resolution optical satellite or aerial images collected from across the globe, covering 46 scene categories. Each image is associated with 1 to 13 labels selected from a predefined set of 60 labels. The dataset construction process includes scene sample collection, database quality control and sample diversity improvement, which ensures the high quality and broad representativeness of the data. MLRSNet is particularly suitable for multi-label image classification and retrieval tasks. Combined with existing datasets such as ImageNet, it can effectively address challenges including object rotation, intra-class variation and inter-class similarity, providing a robust benchmark for remote sensing image analysis.