Deep Learning Based Satellite Image Segmentation Analyzes for Coastal Flood Disaster

Rob floods have become a crucial and complex problem that occurs in many coastal areas throughout the world. This disaster occurs due to many factors, including when sea water rises above normal height and enters coastal areas, causing waterlogging in the area, the impact can damage the environment, infrastructure and disrupt the lives of local communities. Medan Belawan District is one of the sub-districts located in the north of the city of Medan, which borders the Malacca Strait to the north, being a location that is very close to the water so it is predicted that it will be affected by the Rob flood disaster. Deep Learning is a technological innovation whose application can be used to improve flood disaster mitigation. This research uses the Computational Neural Network (CNN) algorithm by implementing the U-Net architecture. U-Net is a type of CNN algorithm architecture which is built with an Encoder-Decoder architecture. Semantic segmentation, which is widely used in classifying satellite images, is one of the methods applied in this research. The segmentation process that applies pixel wise classification aims to make the differences in image details visible and the extraction process can be carried out. The steps taken in this research include Data Gathering, Spatial Filtering, Visual Filtering, Annotation, Pre-Processing, Modeling, Post-Processing and Deployment. In a comparative analysis of image segmentation, researchers compared the results of the U-Net and K-means models used, for semantic segmentation with k- means having two pixel quantization techniques: k-Means with k=2 and multi-otsu with k=5. The U-Net model performs segmentation by utilizing an artificial neural network that has been previously trained to identify and map pixel labels in images.