Optimization Method for Airport Runway Detection Based on U -Net Network

This Runway identification is crucial for enhancing aviation safety, optimizing military operations, and adapting to complex environments. Due to limited data availability, runway extraction still primarily relies on traditional methods. However, the high accuracy and versatility of convolutional neural networks in optical image segmentation tasks present new opportunities in this field. To address this, we propose an airport runway segmentation algorithm based on the U-net network, leveraging deep learning techniques for accurate runway extraction. This algorithm combines transfer learning and focal loss with the U-net network to efficiently and accurately extract runway regions. By utilizing the knowledge of pre-trained models, transfer learning improves training efficiency, model performance, and generalization, which is particularly beneficial when data resources are limited. The introduction of focal loss enhances the model's ability to recognize minority classes (such as runways) and improves overall model performance, which is critical for the segmentation of connected structures like airport runways. Comparative experiments have verified that the proposed algorithm successfully improves detection accuracy in scenarios with limited data.