Water Body Segmentation of Satellite Images using IR U-Net

Water bodies segmentation of satellite images is a very difficult task to perform. In research and development, the ability to identify water bodies using satellite imagery is essential. It can be used to predict natural disasters and identify major crises like drought and flood conditions, among other things, Applications that concentrated on identifying water bodies that are going extinct in the wild are limited. Our present work mainly focuses on Inception-Resnet-v2 U-Net (IR U-Net) segmentation model used to predict water bodies. The images are taken from sentinel-2 satellite which contains water body images and its masks, respectively. In this work, the IR U-net model is used along with TensorFlow to detect the water bodies at the time of natural calamities. Nadam optimizer is used to decrease the losses. The IR U-Net model guarantees the performance accuracy of 90.57for edge detection, blurred image detection, and images with very low resolution. This proposed IR U-Net model is highly robust.