MLGFN: a multilevel gated fusion network for building extraction

Optical imagery and Synthetic Aperture Radar (SAR) imagery, as two important remote sensing data sources, provide complementary information expressions for the same object due to differences in their imaging mechanisms. Effectively integrating multimodal remote sensing imagery is crucial for improving building extraction accuracy. This paper proposes a Multi-Level Gated Fusion Network (MLGFN), which employs a dual branch encoding structure to process optical and SAR images separately. Two novel modules are designed: the Attention Mechanism-Based Gated Fusion Module (AMGFM) for adaptively fusing multimodal features, and the Shallow-Deep Attention Fusion Module (SDAFM) for integrating multi-level semantic information to enhance boundary localization capabilities. Experimental results show that the proposed method outperforms mainstream comparison methods in building extraction tasks, and ablation experiments further validate the effectiveness of each module.