I would like you to replicate the methodology and results as described in the following paper: "Deep-Learning Integration of CNN–Transformer and U-Net for Bi-Temporal SAR Flash-Flood Detection (Noori et al., 2025)" OR you can suggest for me another good Method.
1. Data Provided You will work with the following resources for the Jeddah region:
Two Sentinel-1 images from Copernicus Dataspace:
Pre-flood: 17 November 2022
Post-flood: 29 November 2022
Satellite: Sentinel-1A
Mode: IW (Interferometric Wide Swath)
Polarization: VV or VH (choose the one most suitable for flood mapping)
Product type: GRD (Ground Range Detected)
A 10 m DEM (GeoTIFF format)
A shapefile for Jeddah city
2. Required Workflow A. Preprocessing Complete all preprocessing steps as described in the paper (orbit file application, thermal noise removal, radiometric calibration, speckle filtering, terrain correction, clipping to Jeddah, etc).
Ensure all images are terrain-corrected and co-registered using the provided DEM.
B. Model Implementation Build and train the deep learning model for flood detection (including the bi-temporal input and the hybrid CNN–Transformer or U-Net baseline, as per the paper).
If you have access to the S1GFloods dataset (or similar), you can use it for reference or benchmarking. (https://drive.google.com/file/d/1bm_sFfJ05Fryj6Ib1niIidOywljzIEgo/view?usp=sharing)).
C. Output Provide all outputs in the same format as in the paper (flood maps, binary masks, performance metrics — F1-score, mIoU, etc).
Please organize the outputs for ease of interpretation and comparison with the results from the publication.
D. Documentation Please document all processing and modeling steps, with the parameters/settings used at each stage.
If you need any further clarification, please let me know.
"I would also like to receive the full code (scripts/notebooks) used in the workflow, as well as a brief final report summarizing the results and insights." Thank you!
