Enhanced UAV Autonomous Landing Through YOLOv10-Based Marker Detection with Lightweight and Context-Aware Network
Mengming Wu, Xingguo Song, Xu Fang, Xiao Dong Han · 2025
Autonomous landing of unmanned aerial vehicles (UAVs) under challenging conditions, such as urban tunnel disaster environments, represents a critical research focus. The limited computational resources on UAVs necessitate algorithms that balance high accuracy with computational efficiency. This paper presents UAV-YOLOv10, an optimized landing marker detection algorithm based on an improved YOLOv10 framework. We introduce a Faster Block to enhance the backbone network, significantly reducing model parameters without compromising detection accuracy, making it suitable for deployment on resource-constrained devices. A Dilated Re-Param Block (DRB) embedded in the neck network constructs the C2f-DRB module, leveraging large-kernel and dilated convolutions to expand the receptive field and aggregate contextual information from broader regions. Additionally, an SPPF-SE module is incorporated to enhance selective feature responses, minimizing interference from complex environmental factors. Experimental results demonstrate that UAV-YOLOv10 improves detection accuracy by 3.4%, recall by 0.3%, and [email protected] by 2.3%, while reducing model parameters by 15.5%. Simulation experiments conducted in Gazebo validate the algorithm's ability to reliably, swiftly, and accurately detect landing markers, ensuring successful UAV landing tasks.