Interactive Cloud Detection Based on Dynamic Mode Decomposition
Yilong Li, Lu Yang, Xiaobin Qiu, Chongke Bi · 2024
Cloud detection can address data quality and analytical accuracy issues caused by cloud cover in satellite imagery, significantly enhancing image usability. Existing cloud detection methods still have the problem of detecting a certain amount of ground and sea surface as clouds, as well as some cloud regions that are missed, resulting in a decrease in the accuracy of cloud detection. Aiming at the above problems, an interactive cloud detection method based on Dynamic Mode Decomposition is proposed. Based on the fact that clouds are constantly moving, the dynamic mode decomposition can be utilized to detect the characteristics of moving objects, and the dynamic clouds are detected, avoiding the detection of the ground and sea surface as clouds. Although static clouds rarely appear, to ensure the robustness of the algorithm, we design an interactive static cloud detection algorithm for the users, which solve the problem of missed detection and further improves the accuracy of cloud detection. To assess the effectiveness of the algorithm, the four different types of clouds of floccus, undulatus, vortex clouds, and stratocumulus are detected, the experimental outcomes demonstrate the effectiveness and robustness of the algorithm.