Fanruo Li

Ground penetrating radar (GPR) is a powerful and non-destructive sensing technique that has become increasingly important for monitoring urban road infrastructure. By enabling rapid and accurate acquisition of subsurface information, GPR supports the scientific management and safe operation of underground utilities. However, as urban subsurface environments become more complex, conventional GPR interpretation methods—often based on manual analysis or rule-based algorithms—struggle to maintain high detection accuracy, robustness, and automation. Recent advances in deep learning have opened new opportunities for intelligent GPR data analysis, significantly improving both the efficiency and reliability of underground object detection. This paper provides a comprehensive review of deep learning applications in GPR-based detection of underground targets within urban road environments. It first summarizes typical GPR signal characteristics and data representations for common subsurface objects, followed by a detailed discussion of deep learning techniques employed for object recognition, localization, and classification. Related studies on subsurface parameter inversion are also reviewed to enhance understanding of target geometry and physical properties. Finally, key challenges and future research directions are outlined. This review aims to offer systematic insights into the integration of deep learning with GPR for intelligent underground sensing and to promote its development in urban infrastructure monitoring.

Ground penetrating radar (GPR), as an efficient non-destructive testing technique, plays a crucial role in the structural condition assessment and defect identification of railway ballast. Typical defects such as mud pumping generally exhibit characteristics in B-scan images including weak reflections, blurred boundaries, and irregular structures, which pose significant challenges for stable detection and precise localization using existing methods that rely primarily on spatial feature modeling. Most current deep learning approaches focus on modeling spatial or temporal information, while lacking effective utilization of frequency-domain features, thereby limiting their discriminative capability under complex electromagnetic environments. To address these issues, this paper proposes a single-stage object detection framework, termed YOLO-DGW, based on time-frequency collaborative modeling. Built upon YOLOv8, the proposed method introduces a structure-aware spatial enhancement module to improve the representation of continuous GPR echo structures. Meanwhile, frequency-domain information is incorporated as a modulation prior to guide spatial feature learning, enhancing the model’s sensitivity to weak reflections and complex-shaped targets. In addition, A-CIoU loss function is designed to improve localization accuracy and stability for defect regions of varying scales. Experimental results demonstrate that YOLO-DGW achieves an F1-score of 63.06% and an AP@0.50 of 62.07%, representing improvements of approximately 7.41% and 2.8%, respectively, over the strongest baseline method. Compared with several mainstream object detection models, the proposed approach exhibits superior performance in both detection accuracy and cross-region generalization capability. These findings indicate that integrating frequency-domain information into spatial feature learning through a modulation mechanism can effectively enhance the model’s ability to discriminate weak-reflection anomalies, providing a novel time-frequency collaborative modeling paradigm for railway GPR defect detection.

Ground-penetrating radar (GPR) has been proven effective for detecting subsurface pipes in a nondestructive way, typically with manual processing and decision-making. However, existing automatic models for segmenting the target hyperbolas often lack generalization across different pipe radii, varying subsurface media, and complex field conditions. This is especially reflected in B-scans with diverse or small-scale hyperbolas, often accompanied by cluttered and irregular noise. In this paper, an automatic preprocessing model is proposed to enhance the interpretation of B-scans under challenging conditions. The model includes a ground reflection removal algorithm (GRRA), the data gravitational force enhancement (DGFE) method, and a global–local thresholding technique consisting of dilation-based local thresholding and segmentation (DLTS). First, a frequency-domain filter based on the fast Fourier transform and a spatial filter are applied to the raw B-scan to remove obstructive ground reflection strips. Owing to the minimal intensity differences among the target hyperbola, multiples, and background, the DGFE approach is introduced to amplify the main body of the hyperbola, distinguishing it from the noise. Finally, the target hyperbola is extracted from the grayscale image by an integrated thresholding approach. The approach initially employs global thresholding to eliminate all information except for part of the hyperbola, followed by DLTS, which uses a dilation operation with local thresholding to fully segment the hyperbola. The proposed model is evaluated on two distinct datasets and compared with several state-of-the-art methods. The results demonstrate its effectiveness, particularly in terms of cross-dataset generalization.

Drainage pipelines face significant threats to underground safety due to external defects. Ground Penetrating Radar (GPR) is a primary tool for detecting such defects from within the pipeline. However, existing methods are limited to single or multiple axial scan lines, which cannot provide the precise spatial coordinates of the defects. To address this limitation, this study introduces a novel GPR-based drainage pipeline inspection robot system integrated with multiple sensors. The system incorporates MEMS-IMU, encoder modules, and ultrasonic ranging modules to control the GPR antenna for axial and circumferential scanning. A novel Cross-Line Fusion of GPR (CLF-GPR) method is introduced to integrate axial and circumferential scan data for the precise localization of external pipeline defects. Laboratory simulations were performed to assess the effectiveness of the proposed technology and method, while its practical applicability was further validated through real-world drainage pipeline inspections. The results demonstrate that the proposed approach achieves axial positioning errors of less than 2.0 cm, spatial angular positioning errors below 2°, and depth coordinate errors within 2.3 cm. These findings indicate that the proposed approach is reliable and has the potential to support the transparency and digitalization of urban underground drainage networks.

This study presents a clutter suppression algorithm for underground mine GPR B-scan profiles, integrating the Sparrow Search Algorithm (SSA), Variational Mode Decomposition (VMD), and image fusion via the Improved Non-Subsampled Shearlet Transform (INSST), aimed at addressing clutter under strict explosion-proof conditions. Initially, the raw B-scan data undergoes preprocessing to derive a grayscale profile. Using SSA, an envelope entropy fitness function is introduced to automatically optimize VMD with respect to the average trajectory of the grayscale profile. Finally, enhanced NSST processing is applied to the synthesized profile using a scale-direction adaptive threshold with an L ₂ norm square constraint and a flexible threshold function. This approach reduces Gibbs artifacts and oversmoothing while retaining valid signals within the threshold range. In simulated profiles, the proposed algorithm demonstrates strong robustness across different noise standard deviations, significantly improving image quality metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), while reducing the Root Mean Square Error (RMSE). Notably, in measured mining profiles, the proposed algorithm outperforms eight other methods by better preserving edge details. Compared to the suboptimal algorithm, the Target-to-Noise Ratio (TNR) improves by 36% in mining fault profile I and by 17% in coal seam floor profile II. Finally, a sensitivity analysis of the proposed algorithm’s parameters was conducted.

The underground detection environment in coal mines is complex, with numerous interference sources. Traditional ground-penetrating radar (GPR) methods suffer from limited detection range, high noise levels, and weak deep signals, making it extremely difficult to accurately identify geological structures without stable feature feedback. During research, it was found that the detection energy of the same target significantly changes with the antenna direction. Based on this phenomenon, this paper proposes a geological radar advanced detection method using spatial scanning. This method overcomes constraints imposed by the underground coal mine environment on detection equipment, enhancing both detection range and accuracy compared to traditional approaches. Experiments using this method revealed pea-shaped response characteristics of planar geological structures in radar images, and the mechanisms behind their formation were analyzed. Additionally, this paper studied the changes in response characteristics under changes in target inclination, providing a basis for understanding the spatial distribution of geological structures. Finally, application experiments in underground coal mine environments explored the practical potential of this method. Results indicate that, compared to drilling data, this method achieves identification accuracies of 91.88%, 90.42%, and 78.72% for the depth and spatial extent of geological structures, providing effective technical support for coal mining operations.

The pipeline ground-penetrating radar stands as an indispensable detection device for ensuring underground space security. A wheeled pipeline robot is deployed to traverse the interior of urban underground drainage pipelines along their central axis. It is subject to influences such as resistance, speed, and human factors, leading to deviations in its posture. A guiding wheel is employed to rectify its roll angle and ensure the precise spatial positioning of defects both inside and outside the pipeline, as detected by the radar antenna. By analyzing its deflection factors and correction trajectories, the intelligent correction control of the pipeline ground-penetrating radar falls into the realm of nonlinear multi-constraint optimization. Consequently, a time-series-based correction angle prediction algorithm is proposed. The application of the long short-term memory (LSTM) deep learning model facilitates the prediction of correction angles and torque for the guiding wheel. This study compares the performance of LSTM with an autoregressive integrated moving average model under identical dataset conditions. The subsequent findings reveal a reduction of 4.11° and 8.25 N·m in mean absolute error, and a decrease of 10.66% and 7.27% in mean squared error for the predicted correction angles and torques, respectively. These outcomes are achieved utilizing the three-channel drainage pipeline ground-penetrating radar device with top antenna operating at 1.2 GHz and left/right antennas at 750 MHz. The LSTM prediction model intelligently corrects its posture. Experimental results demonstrate an average correction speed of 5 s and an average angular error of ±1°. It is verified that the model can correct its attitude in real-time with small errors, thereby enhancing the accuracy of ground-penetrating radar antennas in locating pipeline defects.