Earticle

High-resolution remote sensing image scene classification is a challenging visual task, and this study proposes a remote sensing image scene classification method based on Semantic Multi-Granularity Feature Learning Network (SMGFL-Net). The core idea is to learn global and multi-granularity local features from rearranged intermediate feature mappings, thus eliminating meaningless edges. These features are then fused into the final prediction. Through comparative studies, SMGFL-Net consistently outperforms other peer methods in terms of classification accuracy.