Intelligent Water Body Delineation for Flood Monitoring Using Remote Sensing Data and Deep Learning Algorithms

Floods pose significant threats to lives, infrastructure, and the environment. Flood monitoring is crucial for effective disaster management and mitigation. Traditional methods of flood monitoring have several limitations including manual data collection, limited coverage, and delayed response times. In recent years, advanced remote sensing technology, coupled with the power of deep learning algorithms have greatly enhanced flood monitoring capabilities. This paper presents a comprehensive investigation into the integration of remote sensing data and deep learning algorithms for water body delineation in flood monitoring. A dataset containing 700 Sentinel-2 satellite images was used. It was divided into training, testing, and validation data, with corresponding ground truth masks. Deep learning models, including UNET, SEGNET, DeepLabv3, and RefineNet, were employed for water body delineation tasks. After thorough assessment, RefineN et emerged as the best performing model. It has exceptional accuracy and precision in outlining water bodies. Transfer learning techniques were explored to further improve the RefineN et model performance. Various backbone architectures, including ResNet50, ResNet101, MobileNetV2, and EfficientN etBO, were investigated for their compatibility with RefineN et. Experimental results revealed that EfficientN etBO served as the optimal backbone architecture, significantly enhancing the model's performance. The RefineNet model with EfficientN etBO backbone achieved remarkable results, with an accuracy of 0.9478, F1-score of 0.9162, IOU of 0.8455, and MAP of 0.8708. These findings underscore the potential of integrating remote sensing data and deep learning algorithms for flood monitoring applications. The findings from this research carry substantial importance for disaster management and emergency response efforts using neural networks.