Point-Precise Landing of Quadrotor UAV Based on Improved YOLOv10
Yiran Wang, Longwang Huang, Yongfu Li · 2025
Unmanned Aerial Vehicle (UAV) accurate landing is a hot topic in UAV-research field, and machine vision based method is a main stream in this field. However, the landing tag is usually small in the perspective of the UAV, and its resolution is also low, which has great interference to the landing of UAV. To solve these problems, this paper proposes a point-precise landing scheme based on improved YOLOv10. First, to solve the challenge of small target detection, we add SPD-Conv convolutional module to the backbone network of YOLOv10, which significantly improve the feature extraction capability of the model. At the same time, the MCA attention mechanism is integrated into the YOLOv10 network, which enhances the expression ability of the feature map. Then, to promote small target learning ability of YOLO v10, we use a new evaluation index NWD to calculate the loss. Compared with the basic YOLOv10, the improved YOLOv10 has greatly improved the detection accuracy and learning speed in small targets and low-resolution images. Finally, the improved model is trained and deployed on a actual UAV, and accurate landing is achieved during the experiment.