MSAF-Net: Enhancing Multi-Resolution Object Detection in Remote Sensing via Attention-Guided Feature Fusion

Remote sensing object detection faces persistent challenges from scale diversity, cluttered scenes, and resolution constraints, often leading to suboptimal performance on small or overlapping targets. To mitigate these limitations, we introduce MSAF-Net, a novel architecture designed to improve multi-resolution feature learning through refined attention mechanisms. The model incorporates coordinated spatial and channel-wise attention across network stages to adaptively highlight informative regions and suppress background interference. Furthermore, we develop a scale-aware feature fusion approach that aligns representations from different layers, enabling more accurate detection of objects across a wide range of scales. Experimental evaluations on standard benchmarks, including DOTA and NWPU VHR-10, confirm that MSAF-Net achieves notable gains over state-of-the-art methods in both precision and recall. These findings demonstrate the potential of attention-guided strategies in enhancing object detection for high-resolution remote sensing imagery.