Multi-Branch Auxiliary Pyramid and Gaussian Modeling Optimization for UAV Object Detection with YOLOv11

In the task of small target detection in drone aerial images, there are often issues of missed detection and false positives due to small target sizes and complex backgrounds. This paper proposes a small target detection algorithm for aerial images based on YOLOv11 optimization. First, considering that targets in drone aerial images are mostly small, the MAFPN approach is introduced to replace the original neck network, enabling more robust feature interactions across different spatial scales and improving performance in complex scenes. Secondly, based on uncertainty Gaussian modeling techniques, the Bottleneck and C3k2 modules are restructured and named C3k2-GR, which adaptively optimize low-confidence features and suppress image noise. The experimental results of the improved algorithm on drone datasets demonstrate superior performance compared to other models.