Detection of Road Structural Layer Thickness Using LSTM on Ground Penetrating Radar Data

Ground Penetrating Radar (GPR) technology plays a crucial role in road engineering due to its non-destructive, efficient, and accurate assessment of structural layer thickness. Despite being a common function in commercial GPR systems, layer detection typically involves manual selection followed by automatic tracking, resulting in inefficiencies, subjective biases, and the need for post-processing. This paper presents a real-time automatic identification method utilizing Long Short-Term Memory (LSTM) models to analyze GPR data and identify road structural layer thickness. Initially, the GPR data undergo preprocessing. Subsequently, the LSTM model is designed and trained to effectively learn and recognize the A-Scan data. Validation on multiple real road datasets demonstrates that the LSTM-based method outperforms traditional methods in accuracy and robustness. This approach not only enhances the automation of GPR data processing but also overcomes the limitations of traditional methods in real-time automatic detection, providing reliable technical support for road structural layer detection with promising application potential.