Operational cloud monitoring using ground-based all-sky cameras plays a key role in climate research. This paper explores the transition from object detection to pixel-by-pixel semantic segmentation. A comparative analysis of three neural network architectures was conducted: U-Net, DeepLabV3+ and YOLOv8-seg. Using the open WSISEG-Database dataset allowed us to ensure high-quality ground truth labeling, train models with high generalization ability, and determine the optimal architecture for real-time monitoring systems.
all-sky imagery, semantic segmentation, cloud monitoring, deep neural networks, U-Net, DeepLabV3+, YOLOv8-seg
1. Afonasenko A.V., Popova S.D., Shaleev A.V. Metody segmentacii AllSky izobrazheniy dlya zadachi analiza oblachnosti // Optika atmosfery i okeana. Fizika atmosfery. Materialy XXXI Mezhdunarodnogo Simpoziuma. Federal'noe gosudarstvennoe byudzhetnoe uchrezhdenie nauki Institut optiki atmosfery imeni V.E. Zueva Sibirskogo otdeleniya Rossiyskoy akademii nauk. 2025. S. 1-5. DOI:https://doi.org/10.56820/conferencearticle_685d0aa0a7b5f9.30033797.
2. WSISEG-Database [Elektronnyy resurs]. – Rezhim dostupa: https://github.com/CV-Application/WSISEG-Database (data obrascheniya: 27.03.2026).
3. Ronneberger O., Fischer P., Brox T. U-net: Convolutional networks for biomedical image segmentation //International Conference on Medical image computing and computer-assisted intervention. – Cham: Springer international publishing, 2015. – S. 234-241.
4. Yurtkulu S. C., Şahin Y. H., Unal G. Semantic segmentation with extended DeepLabv3 architecture //2019 27th signal processing and communications applications conference (SIU). – IEEE, 2019. – S. 1-4.
