A feature enhancement and attention fusion network for small object detection in UAV imagery

Unmanned aerial vehicle (UAV) small object detection suffers from issues like small target scale, complex backgrounds, and motion blur, which severely limit detection accuracy and robustness. To address these challenges, we propose an enhanced detection framework based on YOLOv11. The framework integrates three core modules: a Multi-Perception Feature Fusion (MPFF) module for fine-grained local feature extraction, an Amplitude-Aware Linear Attention (MALA) module for efficient global context modeling, and a Stepwise Attention Fusion (SAF) module for harmonizing local details and global context. A dedicated dynamic detection head is also added to boost small target sensitivity. Experiments on the VisDrone dataset show that our model achieves a 6.24% improvement in mAP50, 4.45% in mAP95, and 4.55% in recall rate, outperforming mainstream detectors while maintaining low computational complexity. Evaluations on the CCTSDB-2021 and RSOD datasets further validate its strong generalization capability. The proposed method provides a practical solution for UAV applications such as urban surveillance, agricultural monitoring, and disaster response.