Subset of 300 out of 3000 Prepared Sentinel 2 Scenes for Transfer Learning and Super-sampling.

This dataset contains a random subset of 300 out of 3000 Sentinel 2 scenes prepared for Transfer Learning and Super-sampling. It comes in the form of zipped NumPy arrays in the npz format. The files have are named in the following fashion:

latutide+latitude_decimals_longitude+longitude_decimals_month_of_the_year_for_mosaic. 

Each file contains:

bands.npy: The Sentinel 2 bands in 10m resolution uint10: B02, B03, B04, B08, B05, B06, B07, B8A, B11, B12. The 20m bands have been resampled using bilinear resampling.

nir.npy: B08 Resampled to 20m using average resampling and then resampled to 10m using bilinear. Useful for training super-sampling models.

scl.npy: The Sentinel 2 Scene Classification file. Contains information on cloud cover and land cover.

sincos.npy: Contains the latitude, longitude, and time of capture for each pixel encoded to sine and cosine waves in the [0,1] interval. The is useful when training a model to predict where on the globe an image was captured.

The images can be processed to patches using the buteo toolbox: 

`pip install buteo --upgrade

`import buteo as beo`

`beo.get_patches(beo.raster_to_array("path_to_bands"))`

本数据集从3000景Sentinel 2（Sentinel 2）影像中随机选取300景作为子集，专为迁移学习（Transfer Learning）与超分辨率采样（Super-sampling）任务构建。数据集以npz格式压缩的NumPy数组形式存储。文件命名规则如下：
纬度+纬度小数位数_经度+经度小数位数_影像镶嵌所用月份。

每个压缩包内含以下文件：
1. bands.npy：存储10米分辨率、uint10格式的Sentinel 2波段数据，涵盖B02、B03、B04、B08、B05、B06、B07、B8A、B11、B12共10个波段。其中20米分辨率波段已通过双线性重采样（bilinear resampling）重采样至10米分辨率。
2. nir.npy：对应B08波段，先通过平均重采样（average resampling）重采样至20米分辨率，再经双线性重采样转换为10米分辨率，适用于超分辨率采样模型的训练。
3. scl.npy：Sentinel 2场景分类文件，包含云覆盖与土地覆盖相关属性信息。
4. sincos.npy：存储各像素的纬度、经度及成像时间信息，并将上述信息以正弦、余弦波编码至[0,1]区间，可辅助模型训练以预测影像的拍摄地理位置。

可通过buteo工具箱（buteo toolbox）将影像处理为图像块，具体命令如下：
`pip install buteo --upgrade`
`import buteo as beo`
`beo.get_patches(beo.raster_to_array("path_to_bands"))`