Autonomous Airport Runway Recognition for Fixed-Wing Aircraft Based on YOLOv8

In the scenario of autonomous landing for fixed-wing aircraft using vision as the primary sensor, there is a demand for airport runway recognition. Therefore, this paper introduces three modules—Wavelet Transform Convolutional Layer (WTConv), Context-Guided Network (CGNet), and Lightweight Dynamic Upsampling (DySample)—based on YOLOv8 to improve the model. Experimental results demonstrate that after enhancing the base dataset and applying augmentation, the improved model achieves mAP50 scores of 0.65 and 0.734, respectively, representing improvements of 0.07 and 0.04. The GFLOPs of the improved model decrease to 7.9, and the FPS increases, indicating a reduced computational burden and enhanced real-time performance when processing images. The research presented in this paper reduces the computational load during autonomous landing of fixed-wing aircraft, while improving recognition accuracy and recall rate, making the task safer and more efficient. This work is of great significance in promoting the development of autonomous landing technology. Future efforts will focus on further optimizing the model and exploring multi-sensor information fusion to enhance recognition accuracy and robustness.