LaVIDE: Language-Prompted Satellite Change Detection via Map-Image Alignment
Researchers have developed LaVIDE, a novel framework for satellite image change detection that uses language as an intermediary to bridge the semantic gap between map categories and image details. This approach employs restricted prompt learning to align map semantics with image content and an object-aware embedding enhancement to integrate object-level attributes. Experiments on four benchmarks show LaVIDE significantly outperforms existing methods, improving IoU by 18.4% for multi-class and 5.2% for single-class change detection. AI
IMPACT Advances satellite image analysis for applications like urban planning and disaster assessment.