Using Convolutional Neural Networks for Cloud Detection on VEN?S Images over Multiple Land-Cover Types

In most parts of the electromagnetic spectrum, solar radiation cannot penetrate clouds. Therefore, cloud detection and masking are essential in image preprocessing for observing Earth analyzing its properties. Because clouds vary size, shape, structure, an accurate algorithm is required removing them from area interest. This task usually more challenging over bright surfaces such as exposed sunny deserts or snow than water bodies vegetated surfaces. The overarching goal current study to explore compare performance three Convolutional Neural Network architectures (U-Net, SegNet, DeepLab) detecting VEN?S satellite images. To fulfil this goal, tiles Israel were selected. represent different land-use cover categories, including vegetated, urban, agricultural, arid areas, well bodies, with a special focus on desert Additionally, examines effect various channel inputs, exploring possibilities broader usage these data sources. It was found that among tested architectures, U-Net performs best settings. Its results simple RGB-based dataset indicate potential value any system screening, at least visible spectrum. concluded all outperform cloud-masking by lowering false positive ratio tens percents, should be considered alternative user dealing cloud-corrupted scenes.