Multi-Supervised Feature Fusion Attention Network for Clouds and Shadows Detection

Cloud and cloud shadow detection are essential in remote sensing imagery applications. Few semantic segmentation models were designed specifically for clouds their shadows. Based on the visual distribution characteristics of shadows imagery, this paper provides a multi-supervised feature fusion attention network. We design multi-scale block (FFB) problems caused by complex irregular boundaries The consists convolution (FCB), channel (CAB), spatial (SPA). By convolution, FCB reduces excessive differences between shallow deep maps. CAB focuses global local features through attention. Meanwhile, it fuses maps with non-linear weighting to optimize performance. SPA task-relevant areas With three blocks above, alleviates difficulties fusing features. Additionally, makes network resistant background interference while optimizing boundary detection. Our proposed model designs class (CFAB) increase robustness achieves good performance self-made dataset. This dataset is taken from Google Earth contains several satellites. achieved mean intersection over union (MIoU) 94.10% our dataset, which 0.44% higher than other models. Moreover, shows high generalization capability due its superior prediction results HRC_WHU SPARCS datasets.