Adaptive domain-aware network for airport runway subsurface defect detection

Abstract

Ground-penetrating radar (GPR) is widely used in airport runway subsurface defect detection. However, variability in subsurface environments and operational frequencies of GPR systems across different airports can cause significant discrepancies in radar data, which influence defect assessments. To address this problem, this study proposes a deep learning algorithm named AD-DetNet, which is designed to maintain robust generalization performance across various airports under different radar frequency conditions. The AD-DetNet model integrates domain-specific knowledge pertinent to detecting subsurface defects in airport runways, which is suitable for various airport environments. In addition, the AD-DetNet model focuses on identifying and emphasizing common characteristics across diverse airports. Moreover, the proposed model incorporates unlabeled target-domain data during training and employs domain adaptation techniques to align features from different data domains. The results of extensive experiments demonstrate that the proposed AD-DetNet model can achieve superior generalization performance across numerous real-world airport datasets and can outperform current state-of-the-art object detection algorithms.