GPR-STA: A Style Transfer Algorithm for Enhancing GPR Data in Airport Runway Structural Defect Detection

Airport runways, essential for aircraft safety, are susceptible to subsurface defects like voids, settlement, and cracks due to continuous exposure to high loads and complex environments. Traditional detection methods struggle with the identification of various underground anomalies and lack sufficient algorithmic generalization. This study introduces a novel style transfer algorithm, leveraging Ground Penetrating Radar (GPR) data through generative adversarial networks (GANs), to enhance the detection and generalization capabilities in underground object detection at airports. Our method integrates mask and decoder modules within the network architecture and utilizes adversarial learning to manage variations in underground backgrounds, which are common across different airport environments. This approach effectively minimizes the impact of background disparities on the detection algorithms while maintaining spatial consistency of underground objects. The proposed algorithm significantly outperforms traditional methods, as demonstrated in comparative experiments that involve pre- and post-data enhancement. It not only stabilizes image generation, achieving an average structural similarity index above 0.5, but also improves the average F1 score for multi-object detection by 2%. These advancements highlight the enhanced generalization performance of our detection model across varied airport datasets.