Single Image Cloud Detection via Multi-Image Fusion
This work addresses the challenge of costly annotated data collection for cloud detection in remote sensing, which is important for tasks like semantic segmentation and object detection, but it is incremental as it builds on existing multi-image fusion advances.
The paper tackled the problem of cloud detection in remote sensing imagery by leveraging multi-image fusion to bootstrap single image detection, demonstrating that a network optimized for image quality implicitly learns cloud detection and reduces the need for annotated training data.
Artifacts in imagery captured by remote sensing, such as clouds, snow, and shadows, present challenges for various tasks, including semantic segmentation and object detection. A primary challenge in developing algorithms for identifying such artifacts is the cost of collecting annotated training data. In this work, we explore how recent advances in multi-image fusion can be leveraged to bootstrap single image cloud detection. We demonstrate that a network optimized to estimate image quality also implicitly learns to detect clouds. To support the training and evaluation of our approach, we collect a large dataset of Sentinel-2 images along with a per-pixel semantic labelling for land cover. Through various experiments, we demonstrate that our method reduces the need for annotated training data and improves cloud detection performance.