FloodDetectionNet: U-Net Attention Based Flooded Area Segmentation

The paper proposes a novel FloodDetectionNet model for flood area segmentation. The proposed model incorporates attention gates for semantic segmentation by utilizing advanced image segmentation techniques using a modified U-Net architecture. The attention gates focus on critical features gathered from multiple encoder layers, enhancing the accuracy of identifying flood-affected areas. The U-Net with attention structure captures high-level context as well as complex details, which are augmented by dataset modifications to increase generalization. The use of modular attention gate function, provides an effective solution to effectively segment real-world flooding images within the U-Net framework, allowing for improved flood response and prevention strategies. The proposed model has 85.5% precision and 94.4% recall, accurately segments the flood-affected areas. It has achieved 89.9% F1-score, indicating efficient flood detection and contributing to better image segmentation in the U-Net framework, compared to the previously published work.