Flood Area Detection and Segmentation Using Satellite Image in Bangladesh by U-Net: A Deep Learning Approach

Flooding is one of the globe's most prominent and destructive natural calamities, causing significant economic and sociological damages worldwide. Every year, millions of people around the world are affected by flooding. The prevailing flood detection methods are prone to several challenges, such as inconsistencies and inaccuracy due to data manipulation and accessibility constraints. This research presents an effective method for flood propagation and water level segmentation using U-Net. This study employs a U-Net architecture for satellite image data processing of the Feni district of Bangladesh, which allows determining flooding regions at the pixel level with high accuracy. The proposed model identifies inundated areas effectively by learning geometrical patterns related to flooding and segmenting relevant features in the satellite images. The model under consideration achieved a training accuracy of 93.3% and validation accuracy of 90% with minimal loss, hence proving to be reliable and robust. This will lead to an advanced approach in real-time flood mapping that gives accurate information to enhance emergency response, resource allocation, and disaster mitigation strategies. The proposed model, through the use of deep learning techniques and analysis of satellite imagery data, outperforms the traditional hydrological models in terms of efficiency and accuracy.