Water Body Segmentation for Satellite Images Using U-Net++

Satellite images are important for both m onitoring and managing natural resources. The ability to identify and manage water resources is made possible by the segmentation of water bodies in satellite data. In this study, U-Net++ (Nested U-Net) model was used to separate water bodies in satellite data. The dataset for the project was collected using USGS Earth Explorer and QGIS, and it was divided into 20% for testing and 80% for training. After 70 cycles of training, the U-Net++ model had an accuracy of 97.66%. The U-Net++ model builds on the original U-Net model, which has been widely used for segmentation tasks. The U-Net++ model incorporates skip connections and dense connections to improve model performance. This study's ability to segment the water body opens up a lot of possibilities for controlling and monitoring water supplies, among other things. The accuracy reached with the U-Net++ model demonstrates its capacity for accurate water body segmentation in satellite pictures.