Cloud and Cloud Shadow Segmentation for Remote Sensing Imagery Via Filtered Jaccard Loss Function and Parametric Augmentation

Cloud and cloud shadow segmentation are fundamental processes in optical remote sensing image analysis. Current methods for cloud/shadow identification geospatial imagery not as accurate they should, especially the presence of snow haze. This paper presents a deep learning-based framework detection Landsat 8 images. Our method benefits from convolutional neural network, Cloud-Net+ (a modification our previously proposed Cloud-Net \cite{myigarss}) that is trained with novel loss function (Filtered Jaccard Loss). The more sensitive to absence foreground objects an penalizes/rewards predicted mask accurately than other common functions. In addition, sunlight direction-aware data augmentation technique developed task extend generalization ability model by expanding existing training sets. combination Cloud-Net+, Filtered Loss function, algorithm delivers superior results on four public datasets. experiments Pascal VOC dataset exemplifies applicability quality network computer vision applications.