Cloud Detection

Cloud detection in satellite imagery is crucial for accurate Earth observation, aiming to identify and segment cloudy regions to improve the quality of downstream analyses. Current research heavily utilizes deep learning, employing convolutional neural networks (CNNs), transformers, and hybrid architectures like CNN-Transformer models, often incorporating attention mechanisms and multi-resolution processing to capture cloud features at varying scales. These advancements improve accuracy and efficiency, particularly for high-resolution imagery and diverse cloud types, impacting various applications from weather forecasting and climate modeling to precision agriculture and disaster response. Furthermore, research addresses challenges like limited labeled data through synthetic data generation and weakly-supervised learning techniques, and explores efficient on-board processing for autonomous satellite operation.