Topology-Informed Neural Networks for Flood Detection in Optical and Synthetic Aperture Radar Imagery

Short summary

Researchers combine topological data analysis (TDA) with neural networks to improve satellite-based flood detection across optical and SAR imagery. The approach provides interpretable structural features that outperform black-box vision models, especially under cloud cover. Validated on the SEN12-FLOOD dataset, topology-informed systems achieve more robust detection for safety-critical environmental monitoring.

• •Applies topological data analysis to satellite flood detection, addressing cloud-obscured optical imagery limitations
• •Combines TDA features with ResNet and Vision Transformers to improve robustness and interpretability over black-box approaches
• •Validated on open SEN12-FLOOD dataset; topological descriptors work independently and complement existing architectures

Generated with AI, which can make mistakes.