The process involves:

• Extracting important features from both types of data.
  
  


• Using a special model (called a diffusion U-Net) to reduce noise and improve the accuracy of flood boundaries.
  
  


• Refining the final flood map using another U-Net model for precise, pixel-level classification.

To make the model work better, we applied various data augmentation techniques. We tested the approach on well-known datasets and found it performs better than existing methods, especially in tough conditions. The result is a lightweight, robust, and effective system for real-world flood monitoring.