A Spatially Masked Adaptive Gated Network for multimodal post-flood water extent mapping using SAR and incomplete multispectral data

2025-12-31

Isprs Journal of Photogrammetry and Remote Sensing

AI Summary

This study addresses the challenge of post-flood water body mapping under frequent missingness in multispectral imaging (MSI) data due to cloud cover or acquisition delays. To this end, we propose the Spatial Mask Adaptive Gating Network (SMAGNet), which leverages synthetic aperture radar (SAR) as the primary input and dynamically integrates partially or fully missing MSI data through an adaptive gating mechanism. Built upon a U-Net architecture, SMAGNet incorporates spatial masks and gating units to enable robust multimodal feature fusion. Evaluated on the C2S-MS Floods dataset, SMAGNet consistently outperforms existing methods across varying degrees of MSI missingness and achieves performance comparable to SAR-only models when MSI is entirely absent, thereby significantly enhancing mapping accuracy and model robustness in real-world disaster scenarios.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Problem

• post-flood water extent mapping
• multimodal data fusion
• missing MSI data
• SAR imagery
• flood management

Innovation

• multimodal fusion
• missing data robustness
• adaptive gating
• SAR and MSI integration
• flood water mapping