Data for "Annotated dataset for training cloud segmentation neural net-works using high-resolution satellite remote sensing imagery"

The scarcity of specialized datasets and annotation tools presents a formidable hurdle in advancing high-resolution satellite image cloud segmentation algorithms, underscoring the urgency to explore innovative annotation strategies. Ground truth label data stands as a linchpin in the realm of deep learning frameworks, facilitating the iterative refinement of model parameters to achieve optimal predictive accuracy. Despite the dearth of dedicated datasets and annotation software tailored for high-resolution satellite image cloud segmentation tasks, this study introduces CloudLabel, a semi-automatic annotation technique. Harnessing the methods of region growing and morphological processing, CloudLabel streamlines the annotation process for high-resolution satellite images, thereby addressing the limitations of existing annotation platforms which are not specifically adapted to cloud segmentation. By providing a more adaptable and efficient annotation approach compared to conventional labeling methods, CloudLabel enables precise discrimination between cloud and non-cloud regions in high-resolution satellite imagery. Notably, we have curated a dataset comprising 32,065 images (512*512) for cloud segmentation based on CloudLabel, which significantly contributes to algorithmic advancement and broader applications in remote sensing. The 'data' folder contains 32065 RGB images, while the 'label' contains the corresponding CloudLabel labeling results.