Research on small object detection algorithm for UAV aerial images based on AMF-YOLOv11s

Abstract The problem of low detection accuracy in drone aerial object detection due to factors such as large variations in target scale, a high proportion of small-size targets, and complex backgrounds is addressed by proposing an improved drone aerial object detection algorithm, AMF-YOLOv11s, based on YOLOv11. Firstly, the ADown downsampling modules is introduced to reduce computational complexity. Secondly, the MLCA attention module is added after the SPPF layer to enhance the model’s ability to capture small target features. Then, a four-headed adaptive spatial feature fusion module (FASFF) is introduced in the feature fusion layer to strengthen the scale-invariance of features. Finally, a localization regression loss function with NWD metric is adopted to further improve the detection performance for small targets. Experimental results show that, compared to the baseline algorithm, the AMF-YOLOv11s algorithm achieves improvements of 8.8% and 5.6% in mAP50 and mAP50:95, respectively, reaching 48.2% and 29.1%, with a detection speed of 80.9 frames per second. Additionally, the effectiveness and robustness of the proposed algorithm are further validated on the DIOR and UAPD datasets. The AMF-YOLOv11s algorithm proposed in this paper can effectively balance detection accuracy and real-time performance, providing efficient and reliable technical support for object detection in complex drone scenarios.