GDPNet: Gated Dynamic Prototype Network for Few-Shot Remote Sensing Scene Classification

Few-shot remote sensing scene classification (FSRSC) is hindered by large intraclass variance, interclass similarity, and frequent noise in satellite imagery. To address these challenges, a lightweight framework named GDPNet is proposed, coupling multiscale feature fusion with dynamic prototype selection (DPS). Specifically, a gated cross-stage fusion (GCF) module adaptively aggregates features from a MobileViT backbone via channel-spatial attention, enhancing scale-aware discriminability. A DPS module further introduces category-specific thresholds derived from intraclass statistics, yielding more robust and reliable decision boundaries. Experiments on four benchmarks demonstrate that GDPNet achieves the state-of-the-art performance in both five-way one-shot and five-way five-shot settings, while requiring only 2.2 M parameters. These results highlight GDPNet as a compact and effective solution for real-world FSRSC on edge devices.