Very high resolution aerial photography and annotated land cover data of the Peak District National Park

License Contents of compressed file (zip) from Van der Plas, Geikie, Alexander and Simms, upcoming publication titled <em>Multi-stage semantic segmentation quantifies fragmentation of small habitats at a landscape scale</em> This data set contains the RGB image data and human land cover annotation for 1027 patches of 512 pixels x 512 pixels ( 64 m x 64 m spatial resolution). For more information on how the data can be used, the land cover schema and other details, please see our paper. For code examples of how to use the data, please see the github repository at https://github.com/pdnpa/cnn-land-cover The data is given in two formats: python and tiff. The Python format can be directly loaded by the code in the repository into Pytorch DataLoaders. The tiff format is independent of progamming language and application. This data is released under the CC BY 4.0 license, which means if you use this data set, we ask you to cite along with our paper above. If you use the RGB images, you must acknowledge the following copyright: "Aerial Photography for Great Britain, © Bluesky International Limited and Getmapping Plc [2022]" - README land cover patch data.txt - lc_label_names.json contains mapping from land cover label (integer) to land cover class name - python_format - images_python_all all (train and test) RGB images in .npy format (each of shape (3, 512, 512)) - masks_python_all all (train and test) land cover masks in .npy format (each of shape (512, 512)) - train_test_split_80tiles_2023-03-22-2131.json train/test split in json format - train_test_split_80tiles_2023-03-22-2131.pkl train/test split in pickle format (to be used with the data class in the repository) - tiff_format - images_masks_tiff_train train set only patches, containing both the RGB image (first 3 bands) and the land cover annotation (4th band) (each of shape (4, 512, 512)) - images_masks_tiff_test test set only patches, containing both the RGB image (first 3 bands) and the land cover annotation (4th band) (each of shape (4, 512, 512))

本数据集配套的压缩包（ZIP格式）源自Van der Plas、Geikie、Alexander与Simms即将发表的题为《多阶段语义分割量化景观尺度小型生境破碎化程度》（Multi-stage semantic segmentation quantifies fragmentation of small habitats at a landscape scale）的研究论文。
本数据集包含1027幅512像素×512像素（空间分辨率为64米×64米）的图像块的RGB图像数据与人工土地覆盖标注信息。
有关该数据集的使用方式、土地覆盖分类体系及其他细节，请参阅我们的研究论文；有关数据集使用的代码示例，请参阅GitHub仓库：https://github.com/pdnpa/cnn-land-cover。
本数据集提供两种存储格式：Python格式与TIFF格式。其中Python格式可直接通过仓库中的代码加载至PyTorch（PyTorch）数据加载器（DataLoader）中；TIFF格式不依赖编程语言与应用程序。
本数据集采用CC BY 4.0许可协议进行发布。若您使用本数据集，请引用上述研究论文；若您使用RGB图像，必须注明以下版权声明："Aerial Photography for Great Britain, © Bluesky International Limited and Getmapping Plc [2022]"，即「英国航空摄影 © Bluesky International Limited和Getmapping Plc [2022]」。
数据集文件明细如下：
- README land cover patch data.txt：数据集说明文档
- lc_label_names.json：土地覆盖标签（整数）与土地覆盖类别名称的映射文件
- python_format：Python格式数据目录
- images_python_all：包含全部（训练集与测试集）RGB图像，存储为.npy格式，单幅图像形状为(3, 512, 512)
- masks_python_all：包含全部（训练集与测试集）土地覆盖掩码，存储为.npy格式，单幅掩码形状为(512, 512)
- train_test_split_80tiles_2023-03-22-2131.json：JSON格式的训练集/测试集划分文件
- train_test_split_80tiles_2023-03-22-2131.pkl：Pickle格式的训练集/测试集划分文件，需配合仓库中的数据类使用
- tiff_format：TIFF格式数据目录
- images_masks_tiff_train：仅包含训练集图像块，内置RGB图像（前3个波段）与土地覆盖标注（第4个波段），单幅数据形状为(4, 512, 512)
- images_masks_tiff_test：仅包含测试集图像块，内置RGB图像（前3个波段）与土地覆盖标注（第4个波段），单幅数据形状为(4, 512, 512)