Swati Nema, Shweta B. Thomas
Ground penetrating radar (GPR) has emerged as a reliable and high-performance non-destructive testing (NDT) method for pavement evaluation, due to its rapid, non-invasive, and traffic-compatible operation. In this article, 159 peer-reviewed studies published between 1981 and 2025 are systematically studied to review the application of GPR in pavement evaluation includes pavement layer thickness estimation, compaction monitoring, and subsurface defect identification, including cracks, voids, and moisture-related damage. To extract data from signals, efficient signal processing techniques are required. This review systematically summarises the performance of GPR, forward and inverse modelling, various signal processing techniques such as data editing, background clutter removal, IFFT, and advanced high-resolution algorithms such as MUSIC, ESPRIT, etc. According to the published research, advanced signal processing improves the signal-to-noise ratio, increases the accuracy of layer thickness estimation with errors on the order of ∼5% under favourable circumstances, and boosts defect detectability compared to conventional processing. The integration of Artificial Intelligence (AI) models with signal processing methods is increasing rapidly, such as support vector machines, artificial neural networks, convolutional neural networks, etc. which achieve classification accuracies of >90% for pavement distress identification and recognition of subsurface anomalies. Large-scale deployment remains limited by site variability, scarce benchmark datasets, high costs, and uncertainty in material electromagnetic properties, affecting the reliability and scalability of GPR-based pavement evaluation. This review quantitatively compares machine learning and signal processing methods for pavement evaluation and identifies critical research gaps hindering reliable and cost-effective GPR-based pavement management.