Segmentation of Water Bodies from Satellite Images Using Spatial-Channel Attention UNet

Precise division of aquatic areas is essential for efficient environmental supervision and administration. It enables the precise delineation and tracking of aquatic ecosystems, facilitating the assessment of water quality and the identification of environmental hazards such as pollution and habitat degradation. By offering high-resolution data and enabling thorough coverage of large areas, advanced technologies like satellite images play a crucial role in improving water body segmentation and expanding the capacity to monitor and manage water resources effectively and sustainably. In this study, the focus is on leveraging Sentinel-2 satellite imagery for accurate water body segmentation. The dataset is meticulously curated and comprises Sentinel-2 satellite images. Masks are then created to isolate water bodies from the original images using advanced deep-learning architectures, including UNet, Deeplabs, and SegNet. Furthermore, the performance of UNet is enhanced by incorporating attention mechanisms such as channel attention, spatial attention, and combined spatial and channel attention. These algorithms are used to generate output masks, which are then compared with the existing masks from the dataset. Subsequently, the evaluation was performed using metrics such as accuracy, binary cross- entropy loss, and mean intersection over union, providing comprehensive insights into their efficacy. Notably, the findings highlight the remarkable accuracy achieved by the UNet model enhanced with spatial and channel attention mechanisms, reaching an impressive 96.7%. This notable progress emphasizes the possibility of achieving better accuracy in segmenting water, which in turn aids real-time monitoring and informed decision-making in environmental management.