Lightweight Attention-Enhanced Multi-Scale Detector for Robust Small Object Detection in UAV

Small-object detection in unmanned aerial vehicle (UAV) imagery remains challenging due to limited resolution, complex backgrounds, scale variation, and strict real-time constraints. Existing lightweight detectors often struggle to retain fine details while ensuring efficiency, reducing robustness in UAV applications. This letter proposes a lightweight multi-scale frame work integrating Partial Dilated Convolution (PDC), a Triplet Focus Attention Module (TFAM), a Multi-Scale Feature Fusion (MSFF) branch, and a bidirectional BiFPN. PDC enlarges receptive field diversity while preserving local texture, TFAM jointly enhances spatial, channel, and coordinate attention, and MSFF with BiFPN achieves efficient cross-scale fusion. On VisDrone2019, our model reaches 52.7% mAP50 with 6.01M parameters and 148 FPS, and on HIT-UAV yields 85.2% mAP50 and 155 FPS, surpassing state-of-the-art UAV detectors in accuracy and efficiency. Visualization further verifies robustness under low-light, dense, and scale-varying UAV scenes.