Paper Title
Design and Develop Flood Detection and Susceptibility Mapping using GIS and DL Techniques

Abstract
Flooding is a frequent natural disaster impacting infrastructure and communities, especially in vulnerable regions. This study presents a deep learning-based flood detection approach using the SEN12-FLOOD dataset, which includes coregistered “Sentinel-1 SAR and Sentinel-2” optical imagery with annotated flood masks. The dataset contains 13,200 labeled image pairs from 336 global flood-affected regions. This study evaluate three CNNs—ResNet50, DenseNet121, and EfficientNetB0—using only RGB bands from Sentinel-2. All models were fine-tuned via transfer learning and trained with binary cross-entropy loss. Evaluation using accuracy, precision, recall, and F1-score shows EfficientNetB0 performed best. The results demonstrate the potential of combining satellite data with deep learning for accurate, large-scale flood detection. EfficientNetB0’s efficiency and performance make it suitable for real-time disaster monitoring in data-scarce areas. Keywords - Flood Detection, Mapping, GIS(Geographic Information System), Prediction, Susceptibility, Deep Learning