Ground Penetrating Radar (GPR) is an effective non-destructive detection method, that is frequently utilized in the detection of urban underground defects because of its quick speed, convenient and flexible operation, and high resolution. However, there are some limitations to defect detection using GPR, such as less data, poor data quality, and complexity of data interpretation. In this study, an underground defect detection system based on GPR was established. First, a Simple Linear Iterative Clustering (SLIC)-PHash, a Data Augmentation (DA) optimization algorithm, was created to obtain high-quality datasets. Second, the Convolutional Block Attention Module (CBAM)-YOLOv8, a detection model, was produced for the recognition of defects. This model uses GhostConv and CBAM to create a lighter design that better focuses on target detection and increases efficiency. Finally, a one-click detection system was formed by fusing SLIC-Phsh and CBAM-YOLOv8, which were used for one-click GPR dataset optimization and defect detection. The developed system has the best detection mAP and F1 scores of 90.8% and 88.3%, respectively, compared to several well-known Deep Learning (DL)-based techniques. The results demonstrated that the system proposed in this paper can greatly improve detection efficiency and reduce detection time by achieving a good balance between detection speed and accuracy.

This letter proposes a method based on deep learning for the automatic recognition and localization of underground pipelines using the ground penetrating radar (GPR). First, an automatic recognition model with an average precision (AP) of 0.9256 is proposed and trained based on Faster R-CNN. The feature extraction is optimized by the Attention-guided Context Feature Pyramid Network (ACFPN), and the cascade structure is used to improve the detection frame regression accuracy. Moreover, using Tesseract OCR, a positioning model is developed based on recognition results to obtain the burial and horizontal position of the pipeline. Furthermore, on-site experiments were carried out on real embedded pipes to verify the feasibility and effectiveness of the developed method. The absolute error of the localization data is lower than 11 cm, and the average error ratio is smaller than 12%. Consequently, it is demonstrated that the proposed method is considerably automatic, efficient, and reliable for the recognition and localization of underground pipelines.

Due to the unevenness of ground surface, mismatch between trig interval and sampling speed, or other electromagnetic interferences, traces missing is a quite typical occurrence during the on-ground ground penetrating radar (GPR) testing. Effective reconstruction of GPR missing traces has been regarded a crucial link to improve both the signal-to-noise ratio of raw data and the resolution of GPR imaging. In this paper, we propose a novel deep-learning framework based on the residual feature distillation block U-Net (RFDB-U-Net) to mitigate the transmission loss problem of the conventional U-Net. To be specific, by employing the information distillation network based on the multiple feature extraction connections, RFDB is capable of utilizing the adequate residual information of each layer for feature learning. Moreover, a skip connection is additional patched on the residual units to properly compensate the missing features in the convolution process. In particular, the merging of lightweight U-Net ensures the lightness of RFDB. The outperformance of the proposed framework is verified in detail through the reconstruction accuracy and evaluation metrics in the test of synthetic data, laboratorial data and in-site field data.

Ground-penetrating radar (GPR) is widely used in the detection and positioning of underground facilities. Through the inversion analysis of echo signals of GPR, information such as pipe material, burial depth and location of pipelines can be obtained. Unfortunately, underground pipelines are cylindrical, the ladder approximation method used in traditional forward models produces certain errors. In this study, an accurate and efficient numerical model of GPR forward model in underground pipelines is established using symplectic Euler algorithm, graphics processing unit (GPU) acceleration technology and surface conformal technology. With a Ricker wavelet pulse as the GPR source, the convolution perfectly matched layer (CPML) is incorporated in the symplectic Euler algorithm and shown to be effective to truncate the symplectic Euler computational domain. Through the simulation study of different underground pipeline models, GPR image characteristics of the metal pipeline, plastic pipeline and concrete pipeline filled with air and water are obtained. According to the numerical simulation results, parallel conformal symplectic Euler algorithm effectively reduces the false diffracted waves caused by ladder approximation and improves the computational efficiency of the model in metallic and non-metallic media.

Accurate forward modeling is of great significance for improving the accuracy and speed of inversion. For forward modeling of large sizes and fine structures, numerical accuracy and computational efficiency are not high, due to the stability conditions and the dense grid number. In this paper, the symplectic partitioned Runge‐Kutta (SPRK) method, surface conformal technique, and graphics processor unit (GPU) acceleration technique are combined to establish a precise and efficient numerical model of electromagnetic wave propagation in complex geoelectric structures, with the goal of realizing a refined and efficient calculation of the electromagnetic response of an arbitrarily shaped underground target. The results show that the accuracy and efficiency of ground‐penetrating radar (GPR) forward modeling are greatly improved when using our algorithm. This provides a theoretical basis for accurately interpreting GPR detection data and accurate and efficient forward modeling for the next step of inversion imaging.

In engineering practice, ground penetrating radar (GPR) records are often hindered by clutter resulting from uneven underground media distribution, affecting target signal characteristics and precise positioning. To address this issue, we propose a method combining deep learning preprocessing and reverse time migration (RTM) imaging. Our preprocessing approach introduces a novel deep learning framework for GPR clutter, enhancing the network’s feature-capture capability for target signals through the integration of a contextual feature fusion module (CFFM) and an enhanced spatial attention module (ESAM). The superiority and effectiveness of our algorithm are demonstrated by RTM imaging comparisons using synthetic and laboratory data. The processing of actual road data further confirms the algorithm’s significant potential for practical engineering applications.