Explainable Flood Segmentation on Sentinel-1 SAR Imagery: A Comparative Study of CNN and Transformer Architectures
A new research paper compares convolutional neural network (CNN) and vision transformer architectures for flood segmentation using Sentinel-1 SAR imagery. The study found that SegFormer-b2 generally outperformed U-Net on the ETCI dataset, though the advantage diminished on the Sen1Floods11 dataset. The research also employed explainability techniques to understand model decisions and assess reliability. AI
IMPACT This research provides insights into the effectiveness of different AI architectures for satellite-based flood detection, potentially improving disaster response.