Segmentation of Waterbodies from Otsu’s Preprocessed Satellite Images using VGG-UNet

Automatic monitoring of the water-bodies from Satellite Image (SI) is a common process to monitor water reservoir and its impact on the ecological change. When the necessary image is recorded, it is then examined using a chosen imaging scheme to detect the availability of the water and how it can be used in future for a selected application. This research aims to propose and implement a method for automatic segmentation and evaluation of the water region from the chosen SI. In this work, 2000 images/mask is considered to execute the automatic segmentation using pretrained deeplearning models and the performance is confirmed using the raw- and Otsu’s thresholded-image. The experimental outcome of this study is executed on $512 \times 512 \times 3$ pixel sized images and the achieved result confirms that the VGG19-UNet based approach provides a better outcome compared to other models chosen in this study. Further, the result by the Otsu’s processed image is better compared to the raw-image and in future, it can be considered to other SI examination task.