YOLO-TME: A UAV landing detection algorithm that is suitable for Polar ice Floe base stations

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Environmental Monitoring

Background:

• Polar environments present unique challenges for Unmanned Aerial Vehicle (UAV) navigation due to ice floe drift affecting GPS/RTK signals.
• Existing deep learning models struggle with low-contrast, fog/snow-obscured, and high-elevation polar imagery, hindering critical tasks like UAV return and landing.
• Accurate real-time target detection is crucial for autonomous operations in these extreme conditions.

Purpose of the Study:

• To develop an improved deep learning model for robust target detection in polar environments.
• To enhance the accuracy and real-time performance of UAV-based monitoring systems operating in challenging polar conditions.
• To address the limitations of current methods in handling low-visibility and high-altitude imagery for UAV navigation.

Main Methods:

• Integration of transformer-based convolution (TransConv) into the YOLOv11 architecture to enhance global information modeling.
• Introduction of a mist global feature pyramid network (MistGFPN) for improved small target feature extraction at high altitudes.
• Proposal of an efficient asymmetric detection head (EADH) to boost frames per second (FPS) and real-time detection capabilities.

Main Results:

• The proposed YOLO-TME model demonstrated significant improvements over the original YOLOv11.
• Accuracy increased by 4.2%, recall by 5.8%, average precision by 5.5%, and F1 score by 5.1%.
• The model maintains high detection speed, meeting real-time requirements for polar operations.

Conclusions:

• YOLO-TME offers substantially improved detection accuracy and maintains high detection speed in complex polar environments.
• The model effectively addresses challenges posed by fog, snow, and high altitudes for UAV operations.
• YOLO-TME satisfies the critical real-time landing sign detection requirements for UAVs in polar regions.