Navigating Spatial Insights: Comparative Exploration of Deep Learning Algorithms for Flood Detection using Remote Sensing Data

Detecting floods, which are highly destructive natural disasters, is crucial for minimizing their impact on both life and infrastructure [1]. Timely identification is essential, and satellites play a key role in assessing the extent of damage and pinpointing areas in urgent need of assistance following events such as floods, earthquakes, or wildfires. The "SEN12-FLOOD" dataset has emerged as a vital resource for flood detection, utilizing both synthetic aperture radar (SAR) and optical imaging. However, a significant challenge lies in the clarity of the images, making it difficult to accurately identify the extent of damage. In response, we propose a comprehensive methodology that incorporates data preprocessing, the comparison of multiple models, and efficient labeling techniques to enhance the accuracy of flood detection. The data preprocessing phase involves the removal of empty folders and irrelevant images, streamlining the dataset for optimal analysis. Additionally, we introduce band stacking, combining satellite imagery bands B2, B3, B4, and B8 to improve image clarity, providing a refined input for flood detection models. The methodology extends to model comparison, wherein multiple models are evaluated to identify the most effective one for accurate flood detection.