GMBFormer: An NDVI-Guided Global Memory Bank Transformer for Urban Green-Space Extraction from Ultra-High-Resolution Imagery
Researchers have developed GMBFormer, a new Transformer-based framework designed to improve the extraction of urban green spaces from ultra-high-resolution imagery. This model utilizes Normalized Difference Vegetation Index (NDVI) data as a physics-informed gate to selectively admit vegetation descriptors into a global memory bank. By employing memory-mediated cross-attention for prototype retrieval, GMBFormer aims to overcome the limitations of traditional patch-by-patch analysis and improve semantic reuse across spatially separated areas. AI
IMPACT Enhances remote sensing capabilities for urban planning and environmental monitoring.