资源简介:
--- language: - en tags: - clouds - sentinel-2 - image-segmentation - deep-learning - remote-sensing pretty_name: cloudsen12 --- # cloudsen12 ***``A dataset about clouds from Sentinel-2``*** CloudSEN12 is a LARGE dataset (~1 TB) for cloud semantic understanding that consists of 49,400 image patches (IP) that are evenly spread throughout all continents except Antarctica. Each IP covers 5090 x 5090 meters and contains data from Sentinel-2 levels 1C and 2A, hand-crafted annotations of thick and thin clouds and cloud shadows, Sentinel-1 Synthetic Aperture Radar (SAR), digital elevation model, surface water occurrence, land cover classes, and cloud mask results from six cutting-edge cloud detection algorithms. CloudSEN12 is designed to support both weakly and self-/semi-supervised learning strategies by including three distinct forms of hand-crafted labeling data: high-quality, scribble and no-annotation. For more details on how we created the dataset see our paper: CloudSEN12 - a global dataset for semantic understanding of cloud and cloud shadow in Sentinel-2. **ML-STAC Snippet** ```python import mlstac secret = 'https://huggingface.co/datasets/jfloresf/mlstac-demo/resolve/main/main.json' train_db = mlstac.load(secret, framework='torch', stream=True, device='cpu') ``` <p align="center"> <img src="header.png" /> </p> **Sensor: Sentinel2 - MSI** **ML-STAC Task: TensorToTensor, TensorSegmentation** **Data raw repository: [https://cloudsen12.github.io/](https://cloudsen12.github.io/)** **Dataset discussion: [https://github.com/IPL-UV/ML-STAC/discussions/2](https://github.com/IPL-UV/ML-STAC/discussions/2)** **Review mean score: 5.0** **Split_strategy: random** **Paper: [https://www.nature.com/articles/s41597-022-01878-2](https://www.nature.com/articles/s41597-022-01878-2)** ## Data Providers |Name|Role|URL| | :---: | :---: | :---: | |Image & Signal Processing|['host']|https://isp.uv.es/| |ESA|['producer']|https://www.esa.int/| ## Curators |Name|Organization|URL| | :---: | :---: | :---: | |Jair Flores|OEFA|http://jflores.github.io/| ## Reviewers |Name|Organization|URL|Score| | :---: | :---: | :---: | :---: | |Cesar Aybar|Image & Signal Processing|http://csaybar.github.io/|5| ## Labels |Name|Value| | :---: | :---: | |clear|0| |thick-cloud|1| |thin-cloud|2| |cloud-shadow|3| ## Dimensions ### input |Axis|Name|Description| | :---: | :---: | :---: | |0|C|Spectral bands| |1|H|Height| |2|W|Width| ### target |Axis|Name|Description| | :---: | :---: | :---: | |0|C|Hand-crafted labels| |1|H|Height| |2|W|Width| ## Spectral Bands |Name|Common Name|Description|Center Wavelength|Full Width Half Max|Index| | :---: | :---: | :---: | :---: | :---: | :---: | |B01|coastal aerosol|Band 1 - Coastal aerosol - 60m|443.5|17.0|0| |B02|blue|Band 2 - Blue - 10m|496.5|53.0|1| |B03|green|Band 3 - Green - 10m|560.0|34.0|2| |B04|red|Band 4 - Red - 10m|664.5|29.0|3| |B05|red edge 1|Band 5 - Vegetation red edge 1 - 20m|704.5|13.0|4| |B06|red edge 2|Band 6 - Vegetation red edge 2 - 20m|740.5|13.0|5| |B07|red edge 3|Band 7 - Vegetation red edge 3 - 20m|783.0|18.0|6| |B08|NIR|Band 8 - Near infrared - 10m|840.0|114.0|7| |B8A|red edge 4|Band 8A - Vegetation red edge 4 - 20m|864.5|19.0|8| |B09|water vapor|Band 9 - Water vapor - 60m|945.0|18.0|9| |B10|cirrus|Band 10 - Cirrus - 60m|1375.5|31.0|10| |B11|SWIR 1|Band 11 - Shortwave infrared 1 - 20m|1613.5|89.0|11| |B12|SWIR 2|Band 12 - Shortwave infrared 2 - 20m|2199.5|173.0|12|
--- 语言: - 英语 标签: - 云 - Sentinel-2 - 图像分割 - 深度学习 - 遥感 规范名称:cloudsen12 --- # CloudSEN12 ***``基于哨兵二号(Sentinel-2)的云数据集``*** CloudSEN12是一个超大规模(约1 TB)的云语义理解数据集,包含49400张图像块(Image Patches,IP),除南极洲外均匀分布于全球各大陆。每张图像块覆盖5090×5090米的区域,包含哨兵二号(Sentinel-2)1C级与2A级数据、人工标注的厚云、薄云及云阴影标签、哨兵一号(Sentinel-1)合成孔径雷达(Synthetic Aperture Radar, SAR)数据、数字高程模型、地表水发生频率数据、土地覆盖分类数据,以及6种前沿云检测算法生成的云掩膜结果。 CloudSEN12旨在支持弱监督学习以及自/半监督学习策略,其包含三种不同形式的人工标注数据:高质量标注、涂鸦标注(scribble)以及无标注数据。如需了解数据集构建的更多细节,请参阅我们的论文:《CloudSEN12——面向哨兵二号(Sentinel-2)影像中云和云阴影语义理解的全球数据集》。 **ML-STAC 代码示例** python import mlstac secret = 'https://huggingface.co/datasets/jfloresf/mlstac-demo/resolve/main/main.json' train_db = mlstac.load(secret, framework='torch', stream=True, device='cpu') <p align="center"> <img src="header.png" /> </p> **传感器:Sentinel2 - 多光谱成像仪(Multi-Spectral Instrument, MSI)** **ML-STAC 任务:TensorToTensor、张量分割** **原始数据仓库:[https://cloudsen12.github.io/](https://cloudsen12.github.io/)** **数据集讨论区:[https://github.com/IPL-UV/ML-STAC/discussions/2](https://github.com/IPL-UV/ML-STAC/discussions/2)** **评审平均得分:5.0** **划分策略:随机划分** **论文:[https://www.nature.com/articles/s41597-022-01878-2](https://www.nature.com/articles/s41597-022-01878-2)** ## 数据提供方 |名称|角色|URL| | :---: | :---: | :---: | |图像与信号处理实验室(Image & Signal Processing)|['主办方']|https://isp.uv.es/| |欧空局(ESA)|['出品方']|https://www.esa.int/| ## 数据集策展人 |名称|机构|URL| | :---: | :---: | :---: | |Jair Flores|OEFA|http://jflores.github.io/| ## 评审专家 |名称|机构|URL|得分| | :---: | :---: | :---: | :---: | |Cesar Aybar|图像与信号处理实验室|http://csaybar.github.io/|5| ## 标签 |名称|数值| | :---: | :---: | |晴空(clear)|0| |厚云(thick-cloud)|1| |薄云(thin-cloud)|2| |云阴影(cloud-shadow)|3| ## 数据维度 ### 输入数据 |轴序号|名称|描述| | :---: | :---: | :---: | |0|C|光谱波段| |1|H|高度| |2|W|宽度| ### 目标数据 |轴序号|名称|描述| | :---: | :---: | :---: | |0|C|人工标注标签| |1|H|高度| |2|W|宽度| ## 光谱波段 |波段名称|通用名称|描述|中心波长|半高全宽|索引| | :---: | :---: | :---: | :---: | :---: | :---: | |B01|海岸气溶胶波段|波段1 - 海岸气溶胶波段 - 分辨率60m|443.5nm|17.0nm|0| |B02|蓝光波段|波段2 - 蓝光波段 - 分辨率10m|496.5nm|53.0nm|1| |B03|绿光波段|波段3 - 绿光波段 - 分辨率10m|560.0nm|34.0nm|2| |B04|红光波段|波段4 - 红光波段 - 分辨率10m|664.5nm|29.0nm|3| |B05|植被红边1波段|波段5 - 植被红边1波段 - 分辨率20m|704.5nm|13.0nm|4| |B06|植被红边2波段|波段6 - 植被红边2波段 - 分辨率20m|740.5nm|13.0nm|5| |B07|植被红边3波段|波段7 - 植被红边3波段 - 分辨率20m|783.0nm|18.0nm|6| |B08|近红外波段|波段8 - 近红外波段 - 分辨率10m|840.0nm|114.0nm|7| |B8A|植被红边4波段|波段8A - 植被红边4波段 - 分辨率20m|864.5nm|19.0nm|8| |B09|水汽波段|波段9 - 水汽波段 - 分辨率60m|945.0nm|18.0nm|9| |B10|卷云波段|波段10 - 卷云波段 - 分辨率60m|1375.5nm|31.0nm|10| |B11|短波红外1波段|波段11 - 短波红外1波段 - 分辨率20m|1613.5nm|89.0nm|11| |B12|短波红外2波段|波段12 - 短波红外2波段 - 分辨率20m|2199.5nm|173.0nm|12|