An Empirical Evaluation of Deep Convolutional Neural Networks for Flood Detection in Real-Time

Satellite imagery has improved rescue and damage assessment in natural disasters, particularly floods. However, previous water detection methods using satellite images have been inefficient. This study examined flood events in Kerala and Maharashtra in 2018 using Synthetic Aperture Radar (SAR) and three deep learning models: LinkNet, SegNet, and U-Net with MobileNetV2 as backbone. Results showed that LinkNet had the highest accuracy at 92.90%, followed by MobileNetV2 at 92.40%, and SegNet at 86.28%. These findings validate the prospective of SAR and deep learning models in accurate flood identification. Further research could advance flood detection and response strategies, supporting disaster management efforts. The study suggests that LinkNet is the most accurate model among those tested.