Abstract: In order to solve the problems of small target size, dense targets, missed detection and false detection in UAV aerial images, a small target detection algorithm with improved YOLOv11 was proposed, GR-YOLOv11n. Firstly, an efficient feature extraction and fusion module, GhostNet v2, is introduced to replace C3k2 in the backbone network, which can extract richer and more accurate feature information through its unique deep separable convolution and attention mechanism, and enhance the feature extraction ability of the small object detection model. Secondly, the C2f-RepNCSPFPN module was designed, which optimizes the convolution structure and efficient feature processing process by combining the architectural advantages of C2f and RepNCSPFPN, significantly reduces the computational complexity and parameter quantity of the model, and maintains high detection accuracy. On the public dataset VisDrone2019, The proposed GR-YOLOv11n algorithm achieves a 7.6% improvement in mAP@0.5 and a 5.4% gain in mAP@0.5:0.95 compared to the baseline YOLOv11n. Compared with other mainstream object detection methods, the improved model not only satisfies the lightweight, but also improves the detection accuracy of small targets, and meets the requirements of real-time detection.