The Application of Ground Penetrating Radar in the Detection of Hidden Hazards in the Wenzaobang River Dike in Shanghai

The Application of Ground Penetrating Radar in the Detection of Hidden Hazards in the Wenzaobang River Dike in Shanghai | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Application of Ground Penetrating Radar in the Detection of Hidden Hazards in the Wenzaobang River Dike in Shanghai Fuyu Jiang, Run Han, Likun Gao, Jiong Ni, Junkai Yu, Xiaoyu Xu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6633296/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract In maintaining safe navigation along the Wenzaobang River, the dike structure is subject to wave erosion, which can lead to hazards such as surface subsidence, void formation, tilting of the flood control wall, and localized cracking. This study aims to investigate the distribution of potential hazards within the dike by constructing a geological model of loose and voided soil, based on regional geological conditions, and performing forward modeling simulations. The results reveal that loose soil within the dike generates chaotic waveforms in ground-penetrating radar (GPR) images, while voids manifest as arc-shaped reflections. Using the forward modeling outcomes and the YOLOv11 network structure for recognition, the analysis and interpretation of the GPR data indicate the following potential hazard locations: at station 7 + 150 m to 7 + 170 m (depth 0–2 m), loose soil; at station 6 + 960 m to 6 + 970 m (depth 0–5 m), and at station 6 + 870 m to 6 + 890 m (depth 2–9 m), void formation. This research validates the integration of GPR and numerical simulation for geological hazard detection, demonstrating its applicability in complex environments and providing critical data for the design of dike reinforcement. Earth and environmental sciences/Natural hazards Physical sciences/Engineering Dike YOLOv11 Ground Penetrating Radar Loose Void Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Introduction As a critical component of flood control infrastructure, the structural integrity of dike systems directly impacts the flood prevention safety of coastal cities. Over time, dikes are subjected to various factors such as tidal-ship wave coupling erosion, dynamic load disturbances from vehicles, and groundwater infiltration. These influences can lead to concealed damages, including voids, loose soil, and leakage channels, which pose significant risks to flood safety. The right bank dike of the Wenzaobang River, being artificially constructed, faces challenges at the junctions between the old and new sections during reinforcement and maintenance activities. In some areas, the extent of defects is considerable, creating potential safety hazards, particularly during the flood season. Thus, there is an urgent need for efficient and non-destructive detection methods to identify and locate potential hazards. Traditional geophysical methods are often limited by complex hydrological conditions and insufficient spatial resolution, making it challenging to accurately identify potential hazards 1 – 6 . In contrast, ground-penetrating radar (GPR), known for its high resolution and non-destructive detection capabilities, has emerged as the predominant technology for identifying dike defects. GPR operates based on the contrasting electrical properties and geometric configurations of underground media. By analyzing changes in waveform and electromagnetic field strength during the propagation of electromagnetic waves, GPR determines the spatial position and structural characteristics of underground interfaces or geological bodies 7 – 11 .The performance of GPR is influenced by both the antenna frequency and the electrical properties of the medium. High-frequency antennas provide centimeter-level resolution, but their detection depth is limited. Conversely, low-frequency antennas offer greater penetration but suffer from significantly reduced resolution. Moreover, the homogeneity of the medium plays a crucial role in the stability of the electromagnetic wave propagation path. Inhomogeneous strata can lead to signal scattering and energy attenuation, further complicating detection 12 – 20 . Despite ongoing advancements in GPR technology, challenges such as waveform distortion induced by medium heterogeneity and false anomalies resulting from random noise interference persist. Additionally, manual interpretation of GPR data remains inefficient and susceptible to a high rate of misjudgment. YOLO (You Only Look Once) is a deep learning-based, single-stage object detection algorithm that enables end-to-end detection through a single convolutional neural network 21 – 26 . Its core structure consists of several functional modules: convolutional layers for extracting image features, pooling layers for compressing feature dimensions, fully connected layers for predicting object locations and categories, and a detection layer that outputs the final results. With each version iteration, the performance of the YOLO series algorithms has significantly improved. The YOLO series has found wide application in Ground Penetrating Radar (GPR) image processing. For instance, Wang et al. proposed a method for identifying common underground pipelines in GPR images using an improved version of YOLOv5. This method addresses the challenges posed by the small target sizes and irregular shapes of underground pipelines in GPR images, with modifications to the YOLOv5 algorithm enhancing accuracy and recall rates in pipeline detection 27 . Similarly, Yang et al. incorporated transfer learning with YOLOv3 to develop an automatic recognition and localization method for urban underground targets, achieving accurate identification of these targets 28 . Additionally, Qiu et al. introduced an improved YOLOv5 by replacing FPN + PANet with BiFPN, effectively enhancing the recognition performance of GPR images 29 . In response to the aforementioned issues, this study utilizes the GprMax2D platform, integrating the geological characteristics of the Wenzaobang River, to construct a forward model. The study systematically simulates the GPR response characteristics related to soil looseness and void defects. Moreover, by combining a 400 MHz high-frequency antenna with an 80 MHz low-frequency antenna in a collaborative detection approach, the study achieves comprehensive depth coverage of potential hazards and cross-validation of multi-frequency data. Finally, the study employs the YOLOv11 neural network for automatic detection of abnormal hazards in GPR images, facilitating hazard identification and providing a systematic solution for detecting potential risks in complex environmental dikes. 1. Project overview 1.1 Geographic Location The Wenzaobang River is located in the southern part of Jiading District and the central part of Baoshan District in Shanghai. It stretches from the Suzhou River (also known as the Wusong River) in the west to the Huangpu River (at Wusongkou) in the east, with a total length of 33.46 kilometers. Notably, the section from Wendon Lock to Wusong Bridge, situated in the southern part of Baoshan District, extends for 11.16 kilometers. This segment acts as a tributary of the Huangpu River and is positioned within the tidal zone outside the lock, characterized by irregular semi-diurnal tides. The study area is located along the Wenzaobang River in Baoshan District, Shanghai. The river's surrounding topography is predominantly flat, classified as a coastal plain with slight undulations. The study area is located on the right bank of the Wenzaobang River (Fig. 1 ). Continuous river erosion has caused surface cracks, wall fractures, and ground subsidence (Fig. 2 ). These issues not only affect the stability of the dike but also pose potential threats to navigation safety. Therefore, to ensure the normal operation of navigation in the Wenzaobang River, it is crucial to promptly maintain and repair the dike infrastructure in this area. 1.2 Geological Conditions The soil strata exposed to a burial depth of 26.0 meters at the dike site are primarily composed of Holocene Q4 sediments, which exhibit coastal to shallow marine, coastal, and swamp facies. Additionally, late Pleistocene Q3 sediments are present, displaying estuarine to lacustrine, estuarine to coastal, and coastal to shallow marine facies (Fig. 3 ). Based on the origin, composition, and structure of the foundation soils, these strata are classified into six distinct engineering geological layers. A summary of the formation and characteristics of the site is provided as follows: Miscellaneous fill The material is highly heterogeneous, primarily consisting of stones, broken bricks, concrete blocks, and other construction debris, with a minor inclusion of clayey soil. The surface layer is partially overlaid by a concrete layer approximately 20 cm thick. The combination of construction debris and clayey soil creates a strongly heterogeneous medium, resulting in significant scattering noise in radar images. To effectively distinguish between interference and actual anomalies, multi-directional data fusion is essential. Plain fill Light grayish-yellow in color, primarily composed of clayey soil mixed with silty soil and a small amount of stones. The mixture of clayey and silty soils forms a dielectric transition zone, and its gradient characteristics offer a dynamic background field for the detection of shallow loose soil bodies. Silty Clay with Organic Matter Gray in color, containing mica and organic matter, with occasional thin layers of silty soil mixed in. The high water content and organic matter significantly enhance the dielectric properties, which may obscure signals associated with shallow water anomalies. Silty Clay Ranging from dark green to grass yellow, with iron oxide spots and a slightly glossy cut surface. It exhibits medium dry strength and moderate toughness. The cementing effect of iron oxide creates a homogeneous medium, and its stable electromagnetic background facilitates the identification of mid-layer void diffraction characteristics. Clayey Silt Gray in color, saturated, and loose, containing mica with occasional thin layers of sticky soil. The loose structure induces micro-scale dielectric disturbances, resulting in localized distortion of the radar’s in-phase axis. Clay with Organic Matter Gray in color, saturated, and thixotropic, containing mica and organic matter, with occasional thin layers of sandy or silty soil. The high conductivity and thixotropic nature significantly attenuate electromagnetic wave energy, thereby limiting the radar's ability to detect deep-seated hazards. 2. GPR Response Characteristics of Void and Looseness In practical engineering, distinguishing between voids and loose soil using GPR becomes more challenging due to uncertainties in the underground medium caused by soil heterogeneity, water content variations, and environmental interference. A void is a space or cavity filled entirely with air, while loose soil has increased porosity but no continuous cavity. These two differ in terms of forward modeling parameters. Void soil contains many air-filled pores, resulting in a dielectric constant similar to that of air and very low electrical conductivity. Loose soil, with high porosity, contains some moisture, giving it a dielectric constant and conductivity between those of void and normal soil. Normal soil is compact and may contain water, with higher dielectric constant and conductivity, especially when moist or saline. To achieve accurate interpretation of the measured GPR data in the detection area, this paper constructs two geological models—loose soil and void—using GprMax2D software based on the actual conditions of the dike (Table 1 ). Forward simulations of both potential hazards help identify their radar response characteristics, providing a foundation for precise interpretation of the measured data (Fig. 4 ). Table 1 Forward Simulation Parameters Parameter Void Loose Normal Relative Dielectric Constant 1.0 6 10 Conductivity 0 0.001 0.01 Relative Permeability 1.0 1.0 1.0 As is well known, during the construction of dike projects, issues such as uneven compaction, disorganized backfill materials, or uneven backfilling, as well as erosion by river water or compression from passing vehicles, can lead to loose soil in the dike. This, in turn, causes an increase in porosity and a decrease in strength, which affects the structural stability of the dike. On the GPR image (Fig. 4 a), loose soil typically appears as a disordered, region-specific area of strong reflections, with waveforms exhibiting chaotic and irregular characteristics. Disruptions in the phase axis are observed, reflecting a distinct 'chaotic' wave impedance feature. Due to long-term erosion by river water, soil and water loss occur, leading to the formation of soil voids. Additionally, the presence of large cranes on the dike and frequent crane operations can cause local concrete settlement, particularly under concentrated loads. This results in uneven settlement characteristics in the roadbed on both sides of the road, with significant differences in settlement rates. The central area may develop voids due to stress redistribution, requiring deep learning algorithms for automated identification and localization. GPR images of void areas typically show continuous, co-directional reflection wave groups, with noticeable diffracted waves on both sides (Fig. 4 b). 3. YOLOv11 Neural Network Recognition The newly released YOLOv11 enhances feature extraction by incorporating an attention mechanism and a dual-layer depthwise separable convolution structure. These improvements reduce computational load while preserving detection accuracy. As a result, detection efficiency and stability are significantly enhanced, especially in complex scenarios. This study utilizes the YOLOv11 object detection algorithm. First, GPR images are resized to a consistent size and converted to grayscale for recognition. The anomaly maps primarily identify two types of hazards: loose and void areas. The Labelimg tool is then used to annotate the hazardous regions, generating XML format annotation files that include the coordinates of the bounding boxes and their categories. Subsequently, a dam hazard identification and classification dataset is created, consisting of both a training set and a validation set. Finally, radar image recognition experiments are conducted using the YOLOv11 network structure. Figure 5 presents a portion of the recognition results of ground-penetrating radar images in the study area using YOLOv11. As shown, YOLOv11 effectively locates and identifies common hazards, such as voids and loose areas. The experimental results of the YOLOv11-based radar image detection model for void and loose defect identification (Fig. 6 ) demonstrate significant performance improvements due to architectural enhancements. By incorporating a channel attention mechanism, the model improves feature selection while maintaining computational efficiency through depthwise separable convolutions. Over 200 training iterations, the bounding box regression loss (train/box-loss) decreased substantially from 2.5 to 0.5. At the same time, the classification loss (train/cls-loss) converged from 2.0 to 1.0, showing improved performance in both defect localization and classification. Validation metrics indicated excellent detection performance, with the mean average precision at a loose threshold (mAP50(B), IoU = 50%) reaching 0.8. The recall rate (recall(B)) approached 1.0 as iterations progressed, confirming that the model effectively detects defects across the entire image. These quantitative improvements highlight the optimized architecture's suitability for practical defect detection in radar imaging. Overall, the results validate the model's reliability in accurately identifying voids and loose areas, offering an efficient and robust solution for radar image quality assessment. 4. Methods and Data 4.1 Survey Line Arrangement The detection work was organized according to the field conditions, with five survey lines for both the 400 MHz and 80 MHz GPR, running from west to east (Fig. 7 ). Based on the terrain and landform, the detection team conducted trial detections on-site. For mid-high frequency, a 400 MHz shielded antenna with a window length of 128 ns was selected, while for low frequency, an 80 MHz shielded antenna with a window length of 160 ns was chosen. During data collection, the profile method was used for each survey line. The radar antenna was carefully dragged along the water-facing panel, with real-time recording and display monitoring. Each profile survey line was given a unique file number, and all valid records were considered formal detection results. If random factors impacted the detection, re-measurements were taken on-site. The 80 MHz antenna followed the same data collection method as the 400 MHz antenna to ensure the best stacking effect. To accurately identify the distribution of cavities and loose bodies, the detection team conducted on-site investigations. They recorded notable surface and wall cracks, pipelines, collapses, road surface changes, and other conditions. These records were then matched with the corresponding locations in the radar images for later interpretation. 4.2 Ground Penetrating Radar Data Interpretation The YOLOv11 radar image detection model, trained for voids and loose defects, enables rapid and accurate identification of hazardous areas. Below are explanations of some of the abnormal regions observed in the radar images. Figure 8 (a) shows the 400 MHz ground-penetrating radar analysis for survey line L1 (in the 30–50 m range). The radar results in the black rectangular area reveal poor continuity in the in-phase axis of the radar reflection wave, with diffraction and scattering observed in some local sections. The waveforms are chaotic, and a locally upward-opening parabolic shape indicates loose soil layers. Based on borehole data from ZK52 and JK70, it is speculated that between 0 and 40 m along the survey line, at a depth of 1.8–2.2 m, there is a silty clay layer containing organic matter. The upper layers of miscellaneous and plain fill also show signs of sinking. Due to the area’s proximity to the river and significant erosion, the soil here is relatively loose. Figure 8 (b) presents the 400 MHz ground penetrating radar analysis for the northern side of the dike project along survey line L2 (within the 30 ~ 50 m range). Compared to the radar results in the black rectangular area of Fig. 8 (a), the current area shows better continuity of the radar reflection wave’s in-phase axis, with less pronounced diffraction and scattering phenomena. Although the soil layers still exhibit a sinking trend, the degree of sinking is relatively weaker. Therefore, it can be concluded that the closer the area is to the river, the looser the soil structure becomes, leading to higher water content and lower stability of the dike structure. Figure 9 (a) presents the 80 MHz GPR analysis for survey line L5 (in the 280–300 m range). The radar reflection waves within the black rectangular area are highly chaotic, with a significant increase in energy, poor continuity along the in-phase axis, and radar waves displaying a downward-opening parabolic shape. Based on borehole data from JK54, it can be inferred that this layer contains very little plain fill, and the soil properties of the clayey silt are relatively poor. The soil voids in this area are likely related to compaction during construction. Figure 9 (b) shows the 80 MHz GPR results for survey line L6 (in the 350–380 m range). The results reveal a distinct reflection wave anomaly within the black rectangular area. This anomaly is characterized by a diffracted, arc-shaped reflection wave pattern, with overlapping hyperbolic arcs, indicating multiple void areas generating reflection waves in different directions. These waves overlap as they return to the antenna. Additionally, some regions display long, strip-shaped anomalies. This unusual reflection suggests the presence of underground structures or material changes, particularly underground cavities or voids. 5. Discussion and Analysis Based on the analysis of GPR results at different frequencies, Figs. 10 and 11 present the anomaly distribution map and cross-sectional diagram of the detection area, respectively. By combining the engineering geological data and measured results, the following conclusions can be drawn: At stake numbers 7 + 150 m to 7 + 170 m, within a depth of 0 to 2 m, there are loose soil hazards; at stake numbers 6 + 960 m to 6 + 970 m, within a depth of 0 to 5 m, and at stake numbers 6 + 870 m to 6 + 890 m, within a depth of 2 to 9 m, there are void hazards. The void anomaly area is associated with the large cranes at the port. Over time, the frequent use of these cranes has caused settlement in localized areas of the concrete, particularly where the crane loads are concentrated on both sides of the road. This has led to rapid settlement of the roadbed on either side. Due to the uneven nature of the settlement process, a void or cavity may have formed in the central part of the crane operating area, where the gap beneath the concrete was neither filled nor supported. This has resulted in the abnormal reflection waves detected by the GPR. 6. Conclusion Based on the characteristics of the dike project, GPR detection and comprehensive analysis have been utilized to identify the types of hazards, their burial depths, and extents. Loose soil hazards were detected in the horizontal section between pile numbers 7 + 150 m and 7 + 170 m, at depths ranging from 0 to 2 meters. Void hazards were found in the horizontal sections between pile numbers 6 + 960 m and 6 + 970 m, at depths of 0 to 5 meters, and between pile numbers 6 + 870 m and 6 + 890 m, at depths of 2 to 9 meters. The detection results align closely with the adverse geological phenomena identified during the on-site investigation and the stratigraphic features revealed by drilling. These findings indicate that the dike is affected by loose soil and void hazards. Specifically, on the river-facing side, loose dike material with a higher moisture content was found, while near the crane area, voids and potential collapse hazards were detected. These results provide crucial data for the development of the subsequent remediation plan. Declarations Competing interests The authors declare no competing interests. Author Contribution Conceptualization, R.H.; methodology, F.J.; software, L.G.; validation, J.Y.; formal analysis, J.N.; investigation, X.X.; resources, R.H.; data curation, F.J.; writing—original draft preparation, R.H.; writing—review and editing, R.H.; visualization, F.J.; supervision, X.X.; project administration, F.J.; funding acquisition, F.J. All authors have read and agreed to the published version of the manuscript. Acknowledgments This work was supported by the National Key Research and Development Program of China (Grant No. 2021YFC3000103), the National Natural Science Foundation of China (Grant No. 41504081). Data Availability The data presented in this study are available on request from the corresponding author. The data are not publicly available due to institutional policies. References Liu, X. et al. Accurate gain method for ground-penetrating radar signals based on stationary wavelet packet transform. Journal of Applied Geophysics 228 , 105473, doi:https://doi.org/10.1016/j.jappgeo.2024.105473 (2024). Eskandari Torbaghan, M. et al. Automated detection of cracks in roads using ground penetrating radar. Journal of Applied Geophysics 179 , 104118, doi:https://doi.org/10.1016/j.jappgeo.2020.104118 (2020). Zhang, Y. et al. Confidence level-based size estimation of internal crack using multi-trace ground penetrating radar. Construction and Building Materials. 474 , 141124-141124，doi:https://doi.org/10.1016/j.conbuildmat.2025.141124 (2025). Hunziker, J., Meles, G. & Linde, N. Crosshole ground-penetrating radar full-waveform inversion by combining optimal-transport and least-squares distances. Journal of Applied Geophysics 237 , 105655, doi:https://doi.org/10.1016/j.jappgeo.2025.105655 (2025). Da, H. Deep Learning–Based Framework for Bridge Deck Condition Assessment Using Ground Penetrating Radar. Journal of Structural Design and Construction Practice ,doi: https://doi.org/10.1061/JSDCCC.SCENG-1686 30 (2025). Feng, D., Fang, Z., Wang, X. & Ding, S. An efficient footprint-guided finite domain algorithm for common offset ground penetrating radar forward modeling. Journal of Applied Geophysics 233 , 105617, doi:https://doi.org/10.1016/j.jappgeo.2024.105617 (2025). Ruan, W., Liu, H., Dong, H. & Liu, B. Enhanced Ground-Penetrating radar methods for horizon and dielectric characterization of Luvisols. Computers and Electronics in Agriculture ,doi: https://doi.org/10.1016/j.compag.2025.110296 234 , 110296-110296 (2025). Zatar, W., Nguyen, T. T. & Nguyen, H. Environmental effects on condition assessments of concrete structures with ground penetrating radar. Journal of Applied Geophysics 203 , 104713, doi:https://doi.org/10.1016/j.jappgeo.2022.104713 (2022). Dahunsi, J., Pathirana, S., Cheema, M., Krishnapillai, M. & Galagedara, L. Estimating soil hydraulic conductivity from time-lapse ground-penetrating radar data in podzolic soils using the green-ampt model. Journal of Hydrology ,doi: https://doi.org/10.2139/ssrn.4939048 657 , 133059-133059 (2025). Vasques, G. M. et al. Ground penetrating radar (GPR) models of the regolith and water reservoir of an underground dam in the Brazilian semiarid region. Journal of Applied Geophysics 206 , 104797, doi:https://doi.org/10.1016/j.jappgeo.2022.104797 (2022). Pashoutani, S. & Zhu, J. Ground Penetrating Radar Data Processing for Concrete Bridge Deck Evaluation. Journal of Bridge Engineering ,doi: .http://dx.doi.org/10.1061/(asce)be.1943-5592.0001566 25, (2020). Li, K., Xie, X., Chen, H., Zhou, B. & Huang, C. Ground penetrating radar detection of steel fiber reinforced composite linings in shield tunnels: Experimental and field studies. Frontiers of Structural and Civil Engineering ,doi: https://doi.org/10.1007/s11709-025-1165-8 1-15 (2025). Bigman, D. P. & Day, D. J. Ground penetrating radar inspection of a large concrete spillway: A case-study using SFCW GPR at a hydroelectric dam. Case Studies in Construction Materials ,doi: .http://dx.doi.org/10.1016/j.cscm.2022.e00975 16 , (2022). Alarifi, S. S., Ibrahim, E. & Kahtany, K. A. Ground-penetrating radar imaging of subsurface karst features controlling water leakage across Wadi Namar dam, south Riyadh, Saudi Arabia. Open Geosciences ,doi: https://doi.org/10.1515/geo-2025-0781 17 , 20250781-20250781 (2025). Bakker, M. A. J., Maljers, D. & Weerts, H. J. T. Ground-penetrating radar profiling on embanked floodplains. Netherlands Journal of Geosciences-Geologie En Mijnbouw ,doi: .http://dx.doi.org/10.1017/s0016774600021314 86 , 55-61, (2007). Lee, S., Lee, J. H., Joh, C. & Cho, K. Improved delay-and-sum algorithm for ground-penetrating radar imaging considering refraction in multi-layered media. Structures ,doi: https://doi.org/10.2139/ssrn.4910087 75 , 108700-108700 (2025). Gernant, C. et al. Insights into the sea-level history of the South Shetland Islands from ground penetrating radar on Livingston Island, Antarctica. Quaternary Science Reviews ,doi: https://doi.org/10.1130/abs/2020am-356922 359 , 109363-109363 (2025). Aksoy, A., Yigit, E., Caglayan, A. B. & Ozturk, M. Interpretation and analysis of scattering in steel fiber alignment in concrete: The H alpha decomposition method using fully polarized B-scan ground penetrating radar (GPR) image data. Journal of Building Engineering ,doi: https://doi.org/10.1016/j.jobe.2025.112472 105 , 112472-112472 (2025). Coleman, Z. W. & Schindler, A. K. Investigation of ground-penetrating radar defect detection capabilities, influence of moisture content, and optimal data collection orientation in condition assessments of concrete bridge decks. Journal of Applied Geophysics 202 , 104655, doi:https://doi.org/10.1016/j.jappgeo.2022.104655 (2022). He, W., Hao, T., Ke, H., Zheng, W. & Lin, K. Joint time-frequency analysis of ground penetrating radar data based on variational mode decomposition. Journal of Applied Geophysics 181 , 104146, doi:https://doi.org/10.1016/j.jappgeo.2020.104146 (2020). Ayachi, R. et al. Assessing YOLO models for real-time object detection in urban environments for advanced driver-assistance systems (ADAS). Alexandria Engineering Journal 123 , 530-549, doi:https://doi.org/10.1016/j.aej.2025.03.077 (2025). Yeom, J. et al. Automatic Detection of Borehole Breakout for Image Logs using YOLO algorithm. Geoenergy Science and Engineering , 213925, doi:https://doi.org/10.1016/j.geoen.2025.213925 (2025). Akdoğan, C., Özer, T. & Oğuz, Y. PP-YOLO: Deep learning based detection model to detect apple and cherry trees in orchard based on Histogram and Wavelet preprocessing techniques. Computers and Electronics in Agriculture 232 , 110052, doi:https://doi.org/10.1016/j.compag.2025.110052 (2025). Hu, W. et al. Robust crack detection in complex slab track scenarios using STC-YOLO and synthetic data with highly simulated modeling. Automation in Construction 175 , 106219, doi:https://doi.org/10.1016/j.autcon.2025.106219 (2025). Cai, Z., Zhang, Y., Li, J., Zhang, J. & Li, X. Synchronous detection of internal and external defects of citrus by structured-illumination reflectance imaging coupling with improved YOLO v7. Postharvest Biology and Technology 227 , 113576, doi:https://doi.org/10.1016/j.postharvbio.2025.113576 (2025). Song, B. et al. YOLO-ELWNet: A lightweight object detection network. Neurocomputing 636 , 129904, doi:https://doi.org/10.1016/j.neucom.2025.129904 (2025). Wang H Q,. et al. Improved YOLOv5 for common underground pipeline detection using ground-penetrating radar. Chinese Journal of Geophysics . 2024, 67: 3588-3604. .28 Yang B S. et al. Real-time detection method for underground targets using vehicle-mounted ground-penetrating radar. Acta Geodaetica et Cartographica Sinica . 2020, 49: 874-882. Zhi, Q. et al. Application of an Improved YOLOv5 Algorithm in Real-Time Detection of Foreign Objects by Ground Penetrating Radar. Remote Sensing. 14 , 1895-1895 (2022). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6633296","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":477101511,"identity":"1083ef21-a85d-4348-bb28-67c61e4d70e2","order_by":0,"name":"Fuyu Jiang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1UlEQVRIiWNgGAWjYBACxmYQacMgZyABYrARrSWNwZh4LRCQxpC4gWgtzO3Mzx5+STicvl26x4DhQ9lhBv7ZDYQcxmZuLJNwOHfnnDMGjDPOHWaQuHOAkBYGM2nJH4dzN9zIMWDmbTvMYCCRQEgL+zdpCaDDDEBa/hKnhcdM8kPC4QSwFkYitZRJMySkG264kVZwsOdcOo/EDQJaDPuPb5P8kWAtb3AjeeODH2XWcvwzCGlpAAY0DwM4QhkOADEPfvVAIA9y3A+GOoIKR8EoGAWjYAQDAHm9QnQlRVcSAAAAAElFTkSuQmCC","orcid":"","institution":"Hohai University","correspondingAuthor":true,"prefix":"","firstName":"Fuyu","middleName":"","lastName":"Jiang","suffix":""},{"id":477101512,"identity":"b5b27c1e-fb24-47fe-8b2a-86752321a64a","order_by":1,"name":"Run Han","email":"","orcid":"","institution":"Hohai University","correspondingAuthor":false,"prefix":"","firstName":"Run","middleName":"","lastName":"Han","suffix":""},{"id":477101513,"identity":"71af8025-6935-46f0-96a7-f3e681ca88da","order_by":2,"name":"Likun Gao","email":"","orcid":"","institution":"The first geological brigade of Jiangsu geological bureau","correspondingAuthor":false,"prefix":"","firstName":"Likun","middleName":"","lastName":"Gao","suffix":""},{"id":477101514,"identity":"e60fc559-9abe-4d52-be92-605602d938c2","order_by":3,"name":"Jiong Ni","email":"","orcid":"","institution":"Hohai University","correspondingAuthor":false,"prefix":"","firstName":"Jiong","middleName":"","lastName":"Ni","suffix":""},{"id":477101515,"identity":"7f37f0b8-55e8-4963-83e7-0cdad3ec4393","order_by":4,"name":"Junkai Yu","email":"","orcid":"","institution":"Hohai University","correspondingAuthor":false,"prefix":"","firstName":"Junkai","middleName":"","lastName":"Yu","suffix":""},{"id":477101516,"identity":"16c386c9-7fcd-4eff-9c50-51b435c4ee66","order_by":5,"name":"Xiaoyu Xu","email":"","orcid":"","institution":"Hohai University","correspondingAuthor":false,"prefix":"","firstName":"Xiaoyu","middleName":"","lastName":"Xu","suffix":""}],"badges":[],"createdAt":"2025-05-10 07:53:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6633296/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6633296/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87575221,"identity":"559b6222-3baf-4478-bad1-c6c79b41fd3f","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":452086,"visible":true,"origin":"","legend":"\u003cp\u003eDescription of the study area.\u003c/p\u003e","description":"","filename":"image1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/ac8be8ef945ed901fd3a59ff.jpeg"},{"id":87577213,"identity":"cb3135a5-d385-400e-abb9-854cfdd8c867","added_by":"auto","created_at":"2025-07-25 11:52:16","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":343007,"visible":true,"origin":"","legend":"\u003cp\u003ePhotographs of on-site anomalies.\u003c/p\u003e\n\u003cp\u003e(a) Potential void formation hazard (b) Loose hazard\u003c/p\u003e\n\u003cp\u003e(c) Cracking hazard (d) Settlement hazard\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/1d2aa36b1ae98860bd75415a.png"},{"id":87576610,"identity":"1660b64a-36f2-44b9-9e67-ade4e1a28a0b","added_by":"auto","created_at":"2025-07-25 11:44:16","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":189119,"visible":true,"origin":"","legend":"\u003cp\u003eEngineering geological profile of the study area\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/3b70fc9bbf72b71d214deced.png"},{"id":87575226,"identity":"f8a05d66-ff51-4900-ae21-1956813be9ec","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":96565,"visible":true,"origin":"","legend":"\u003cp\u003eForward model diagram\u003c/p\u003e\n\u003cp\u003e(a) Loose model (b) Void model\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/4ea88495753e90fb030016da.png"},{"id":87575223,"identity":"9f8998c5-34b0-4ab4-ae2d-8cd3fff12e7f","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":344566,"visible":true,"origin":"","legend":"\u003cp\u003eYOLOv11 Recognition Results on GPR Images in the Study Area\u003c/p\u003e","description":"","filename":"image5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/3067122cca807f3788aac003.jpeg"},{"id":87575238,"identity":"3ed830b3-b11a-4d7f-9a43-6e8c54a54064","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":227325,"visible":true,"origin":"","legend":"\u003cp\u003eYOLOv11 Training Performance\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/7193927724cac4569efb41d0.png"},{"id":87576612,"identity":"b861e610-340c-4e84-857b-1e3b8560b41a","added_by":"auto","created_at":"2025-07-25 11:44:16","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":127756,"visible":true,"origin":"","legend":"\u003cp\u003eSurvey Line Layout Diagram\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/8e4ba8384a4ce0077ffd506e.png"},{"id":87575239,"identity":"eb49d32a-5c3a-4881-b588-d2984a454e75","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":232393,"visible":true,"origin":"","legend":"\u003cp\u003eGPR Analysis of Loose Hazard Results\u003c/p\u003e\n\u003cp\u003e(a) 400 MHz GPR Analysis of Survey Line L1\u003c/p\u003e\n\u003cp\u003e(b) 400 MHz GPR Analysis of Survey Line L2\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/5dc7bcfd9550c8e23697554b.png"},{"id":87575247,"identity":"48febea6-cd72-4792-945a-dd9eaad0322b","added_by":"auto","created_at":"2025-07-25 11:36:17","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":205014,"visible":true,"origin":"","legend":"\u003cp\u003eGPR Analysis of Void Hazard Results\u003c/p\u003e\n\u003cp\u003e(a) 80 MHz GPR Analysis of Survey Line L5\u003c/p\u003e\n\u003cp\u003e(b) 80 MHz GPR Analysis of Survey Line L6\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/ea743b1747490c286e0f5841.png"},{"id":87575231,"identity":"c1bb7a09-8fd0-4d8b-9bfd-9247b695bd38","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":108178,"visible":true,"origin":"","legend":"\u003cp\u003eHazard Distribution Plan Map\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/f291db7c94f826f4ee37f816.png"},{"id":87575236,"identity":"66980fbd-6688-4735-bdf3-9e388f088969","added_by":"auto","created_at":"2025-07-25 11:36:16","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":143897,"visible":true,"origin":"","legend":"\u003cp\u003eHazard Distribution Section Map\u003c/p\u003e\n\u003cp\u003e(a) Section 1 (b) Section 2 (c) Section 3 (d) Section 4\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/8b97a26953d16cb49a601b32.png"},{"id":100901472,"identity":"cfd5f37a-5f8e-46f8-8c61-e57c501f004d","added_by":"auto","created_at":"2026-01-22 14:56:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3101957,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6633296/v1/7af5fe1f-1b9a-4c27-95ee-8f02c1e493b3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"The Application of Ground Penetrating Radar in the Detection of Hidden Hazards in the Wenzaobang River Dike in Shanghai","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAs a critical component of flood control infrastructure, the structural integrity of dike systems directly impacts the flood prevention safety of coastal cities. Over time, dikes are subjected to various factors such as tidal-ship wave coupling erosion, dynamic load disturbances from vehicles, and groundwater infiltration. These influences can lead to concealed damages, including voids, loose soil, and leakage channels, which pose significant risks to flood safety. The right bank dike of the Wenzaobang River, being artificially constructed, faces challenges at the junctions between the old and new sections during reinforcement and maintenance activities. In some areas, the extent of defects is considerable, creating potential safety hazards, particularly during the flood season. Thus, there is an urgent need for efficient and non-destructive detection methods to identify and locate potential hazards.\u003c/p\u003e\u003cp\u003eTraditional geophysical methods are often limited by complex hydrological conditions and insufficient spatial resolution, making it challenging to accurately identify potential hazards \u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. In contrast, ground-penetrating radar (GPR), known for its high resolution and non-destructive detection capabilities, has emerged as the predominant technology for identifying dike defects. GPR operates based on the contrasting electrical properties and geometric configurations of underground media. By analyzing changes in waveform and electromagnetic field strength during the propagation of electromagnetic waves, GPR determines the spatial position and structural characteristics of underground interfaces or geological bodies \u003csup\u003e\u003cspan additionalcitationids=\"CR8 CR9 CR10\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e.The performance of GPR is influenced by both the antenna frequency and the electrical properties of the medium. High-frequency antennas provide centimeter-level resolution, but their detection depth is limited. Conversely, low-frequency antennas offer greater penetration but suffer from significantly reduced resolution. Moreover, the homogeneity of the medium plays a crucial role in the stability of the electromagnetic wave propagation path. Inhomogeneous strata can lead to signal scattering and energy attenuation, further complicating detection \u003csup\u003e\u003cspan additionalcitationids=\"CR13 CR14 CR15 CR16 CR17 CR18 CR19\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Despite ongoing advancements in GPR technology, challenges such as waveform distortion induced by medium heterogeneity and false anomalies resulting from random noise interference persist. Additionally, manual interpretation of GPR data remains inefficient and susceptible to a high rate of misjudgment.\u003c/p\u003e\u003cp\u003eYOLO (You Only Look Once) is a deep learning-based, single-stage object detection algorithm that enables end-to-end detection through a single convolutional neural network \u003csup\u003e\u003cspan additionalcitationids=\"CR22 CR23 CR24 CR25\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Its core structure consists of several functional modules: convolutional layers for extracting image features, pooling layers for compressing feature dimensions, fully connected layers for predicting object locations and categories, and a detection layer that outputs the final results. With each version iteration, the performance of the YOLO series algorithms has significantly improved. The YOLO series has found wide application in Ground Penetrating Radar (GPR) image processing. For instance, Wang et al. proposed a method for identifying common underground pipelines in GPR images using an improved version of YOLOv5. This method addresses the challenges posed by the small target sizes and irregular shapes of underground pipelines in GPR images, with modifications to the YOLOv5 algorithm enhancing accuracy and recall rates in pipeline detection \u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Similarly, Yang et al. incorporated transfer learning with YOLOv3 to develop an automatic recognition and localization method for urban underground targets, achieving accurate identification of these targets \u003csup\u003e28\u003c/sup\u003e. Additionally, Qiu et al. introduced an improved YOLOv5 by replacing FPN\u0026thinsp;+\u0026thinsp;PANet with BiFPN, effectively enhancing the recognition performance of GPR images \u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn response to the aforementioned issues, this study utilizes the GprMax2D platform, integrating the geological characteristics of the Wenzaobang River, to construct a forward model. The study systematically simulates the GPR response characteristics related to soil looseness and void defects. Moreover, by combining a 400 MHz high-frequency antenna with an 80 MHz low-frequency antenna in a collaborative detection approach, the study achieves comprehensive depth coverage of potential hazards and cross-validation of multi-frequency data. Finally, the study employs the YOLOv11 neural network for automatic detection of abnormal hazards in GPR images, facilitating hazard identification and providing a systematic solution for detecting potential risks in complex environmental dikes.\u003c/p\u003e"},{"header":"1. Project overview","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e1.1 Geographic Location\u003c/h2\u003e\n \u003cp\u003eThe Wenzaobang River is located in the southern part of Jiading District and the central part of Baoshan District in Shanghai. It stretches from the Suzhou River (also known as the Wusong River) in the west to the Huangpu River (at Wusongkou) in the east, with a total length of 33.46 kilometers. Notably, the section from Wendon Lock to Wusong Bridge, situated in the southern part of Baoshan District, extends for 11.16 kilometers. This segment acts as a tributary of the Huangpu River and is positioned within the tidal zone outside the lock, characterized by irregular semi-diurnal tides. The study area is located along the Wenzaobang River in Baoshan District, Shanghai. The river\u0026apos;s surrounding topography is predominantly flat, classified as a coastal plain with slight undulations.\u003c/p\u003e\n \u003cp\u003eThe study area is located on the right bank of the Wenzaobang River (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). Continuous river erosion has caused surface cracks, wall fractures, and ground subsidence (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). These issues not only affect the stability of the dike but also pose potential threats to navigation safety. Therefore, to ensure the normal operation of navigation in the Wenzaobang River, it is crucial to promptly maintain and repair the dike infrastructure in this area.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003e1.2 Geological Conditions\u003c/h3\u003e\n\u003cp\u003eThe soil strata exposed to a burial depth of 26.0 meters at the dike site are primarily composed of Holocene Q4 sediments, which exhibit coastal to shallow marine, coastal, and swamp facies. Additionally, late Pleistocene Q3 sediments are present, displaying estuarine to lacustrine, estuarine to coastal, and coastal to shallow marine facies (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). Based on the origin, composition, and structure of the foundation soils, these strata are classified into six distinct engineering geological layers. A summary of the formation and characteristics of the site is provided as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMiscellaneous fill\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe material is highly heterogeneous, primarily consisting of stones, broken bricks, concrete blocks, and other construction debris, with a minor inclusion of clayey soil. The surface layer is partially overlaid by a concrete layer approximately 20 cm thick. The combination of construction debris and clayey soil creates a strongly heterogeneous medium, resulting in significant scattering noise in radar images. To effectively distinguish between interference and actual anomalies, multi-directional data fusion is essential.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePlain fill\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLight grayish-yellow in color, primarily composed of clayey soil mixed with silty soil and a small amount of stones. The mixture of clayey and silty soils forms a dielectric transition zone, and its gradient characteristics offer a dynamic background field for the detection of shallow loose soil bodies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSilty Clay with Organic Matter\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGray in color, containing mica and organic matter, with occasional thin layers of silty soil mixed in. The high water content and organic matter significantly enhance the dielectric properties, which may obscure signals associated with shallow water anomalies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSilty Clay\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRanging from dark green to grass yellow, with iron oxide spots and a slightly glossy cut surface. It exhibits medium dry strength and moderate toughness. The cementing effect of iron oxide creates a homogeneous medium, and its stable electromagnetic background facilitates the identification of mid-layer void diffraction characteristics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClayey Silt\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGray in color, saturated, and loose, containing mica with occasional thin layers of sticky soil. The loose structure induces micro-scale dielectric disturbances, resulting in localized distortion of the radar\u0026rsquo;s in-phase axis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClay with Organic Matter\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGray in color, saturated, and thixotropic, containing mica and organic matter, with occasional thin layers of sandy or silty soil. The high conductivity and thixotropic nature significantly attenuate electromagnetic wave energy, thereby limiting the radar\u0026apos;s ability to detect deep-seated hazards.\u003c/p\u003e"},{"header":"2. GPR Response Characteristics of Void and Looseness","content":"\u003cp\u003eIn practical engineering, distinguishing between voids and loose soil using GPR becomes more challenging due to uncertainties in the underground medium caused by soil heterogeneity, water content variations, and environmental interference. A void is a space or cavity filled entirely with air, while loose soil has increased porosity but no continuous cavity. These two differ in terms of forward modeling parameters. Void soil contains many air-filled pores, resulting in a dielectric constant similar to that of air and very low electrical conductivity. Loose soil, with high porosity, contains some moisture, giving it a dielectric constant and conductivity between those of void and normal soil. Normal soil is compact and may contain water, with higher dielectric constant and conductivity, especially when moist or saline.\u003c/p\u003e\u003cp\u003eTo achieve accurate interpretation of the measured GPR data in the detection area, this paper constructs two geological models\u0026mdash;loose soil and void\u0026mdash;using GprMax2D software based on the actual conditions of the dike (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Forward simulations of both potential hazards help identify their radar response characteristics, providing a foundation for precise interpretation of the measured data (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eForward Simulation Parameters\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVoid\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLoose\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNormal\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRelative Dielectric Constant\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eConductivity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.01\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRelative Permeability\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAs is well known, during the construction of dike projects, issues such as uneven compaction, disorganized backfill materials, or uneven backfilling, as well as erosion by river water or compression from passing vehicles, can lead to loose soil in the dike. This, in turn, causes an increase in porosity and a decrease in strength, which affects the structural stability of the dike. On the GPR image (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea), loose soil typically appears as a disordered, region-specific area of strong reflections, with waveforms exhibiting chaotic and irregular characteristics. Disruptions in the phase axis are observed, reflecting a distinct 'chaotic' wave impedance feature.\u003c/p\u003e\u003cp\u003eDue to long-term erosion by river water, soil and water loss occur, leading to the formation of soil voids. Additionally, the presence of large cranes on the dike and frequent crane operations can cause local concrete settlement, particularly under concentrated loads. This results in uneven settlement characteristics in the roadbed on both sides of the road, with significant differences in settlement rates. The central area may develop voids due to stress redistribution, requiring deep learning algorithms for automated identification and localization. GPR images of void areas typically show continuous, co-directional reflection wave groups, with noticeable diffracted waves on both sides (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb).\u003c/p\u003e"},{"header":"3. YOLOv11 Neural Network Recognition","content":"\u003cp\u003eThe newly released YOLOv11 enhances feature extraction by incorporating an attention mechanism and a dual-layer depthwise separable convolution structure. These improvements reduce computational load while preserving detection accuracy. As a result, detection efficiency and stability are significantly enhanced, especially in complex scenarios. This study utilizes the YOLOv11 object detection algorithm. First, GPR images are resized to a consistent size and converted to grayscale for recognition. The anomaly maps primarily identify two types of hazards: loose and void areas. The Labelimg tool is then used to annotate the hazardous regions, generating XML format annotation files that include the coordinates of the bounding boxes and their categories. Subsequently, a dam hazard identification and classification dataset is created, consisting of both a training set and a validation set. Finally, radar image recognition experiments are conducted using the YOLOv11 network structure.\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents a portion of the recognition results of ground-penetrating radar images in the study area using YOLOv11. As shown, YOLOv11 effectively locates and identifies common hazards, such as voids and loose areas. The experimental results of the YOLOv11-based radar image detection model for void and loose defect identification (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e) demonstrate significant performance improvements due to architectural enhancements. By incorporating a channel attention mechanism, the model improves feature selection while maintaining computational efficiency through depthwise separable convolutions. Over 200 training iterations, the bounding box regression loss (train/box-loss) decreased substantially from 2.5 to 0.5. At the same time, the classification loss (train/cls-loss) converged from 2.0 to 1.0, showing improved performance in both defect localization and classification. Validation metrics indicated excellent detection performance, with the mean average precision at a loose threshold (mAP50(B), IoU\u0026thinsp;=\u0026thinsp;50%) reaching 0.8. The recall rate (recall(B)) approached 1.0 as iterations progressed, confirming that the model effectively detects defects across the entire image. These quantitative improvements highlight the optimized architecture's suitability for practical defect detection in radar imaging. Overall, the results validate the model's reliability in accurately identifying voids and loose areas, offering an efficient and robust solution for radar image quality assessment.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"4. Methods and Data","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003e4.1 Survey Line Arrangement\u003c/h2\u003e\n \u003cp\u003eThe detection work was organized according to the field conditions, with five survey lines for both the 400 MHz and 80 MHz GPR, running from west to east (Fig. \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e). Based on the terrain and landform, the detection team conducted trial detections on-site. For mid-high frequency, a 400 MHz shielded antenna with a window length of 128 ns was selected, while for low frequency, an 80 MHz shielded antenna with a window length of 160 ns was chosen. During data collection, the profile method was used for each survey line. The radar antenna was carefully dragged along the water-facing panel, with real-time recording and display monitoring. Each profile survey line was given a unique file number, and all valid records were considered formal detection results. If random factors impacted the detection, re-measurements were taken on-site. The 80 MHz antenna followed the same data collection method as the 400 MHz antenna to ensure the best stacking effect.\u003c/p\u003e\n \u003cp\u003eTo accurately identify the distribution of cavities and loose bodies, the detection team conducted on-site investigations. They recorded notable surface and wall cracks, pipelines, collapses, road surface changes, and other conditions. These records were then matched with the corresponding locations in the radar images for later interpretation.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003e4.2 Ground Penetrating Radar Data Interpretation\u003c/h3\u003e\n\u003cp\u003eThe YOLOv11 radar image detection model, trained for voids and loose defects, enables rapid and accurate identification of hazardous areas. Below are explanations of some of the abnormal regions observed in the radar images.\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e(a) shows the 400 MHz ground-penetrating radar analysis for survey line L1 (in the 30\u0026ndash;50 m range). The radar results in the black rectangular area reveal poor continuity in the in-phase axis of the radar reflection wave, with diffraction and scattering observed in some local sections. The waveforms are chaotic, and a locally upward-opening parabolic shape indicates loose soil layers. Based on borehole data from ZK52 and JK70, it is speculated that between 0 and 40 m along the survey line, at a depth of 1.8\u0026ndash;2.2 m, there is a silty clay layer containing organic matter. The upper layers of miscellaneous and plain fill also show signs of sinking. Due to the area\u0026rsquo;s proximity to the river and significant erosion, the soil here is relatively loose.\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e(b) presents the 400 MHz ground penetrating radar analysis for the northern side of the dike project along survey line L2 (within the 30\u0026thinsp;~\u0026thinsp;50 m range). Compared to the radar results in the black rectangular area of Fig. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e(a), the current area shows better continuity of the radar reflection wave\u0026rsquo;s in-phase axis, with less pronounced diffraction and scattering phenomena. Although the soil layers still exhibit a sinking trend, the degree of sinking is relatively weaker. Therefore, it can be concluded that the closer the area is to the river, the looser the soil structure becomes, leading to higher water content and lower stability of the dike structure.\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e(a) presents the 80 MHz GPR analysis for survey line L5 (in the 280\u0026ndash;300 m range). The radar reflection waves within the black rectangular area are highly chaotic, with a significant increase in energy, poor continuity along the in-phase axis, and radar waves displaying a downward-opening parabolic shape. Based on borehole data from JK54, it can be inferred that this layer contains very little plain fill, and the soil properties of the clayey silt are relatively poor. The soil voids in this area are likely related to compaction during construction.\u003c/p\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e(b) shows the 80 MHz GPR results for survey line L6 (in the 350\u0026ndash;380 m range). The results reveal a distinct reflection wave anomaly within the black rectangular area. This anomaly is characterized by a diffracted, arc-shaped reflection wave pattern, with overlapping hyperbolic arcs, indicating multiple void areas generating reflection waves in different directions. These waves overlap as they return to the antenna. Additionally, some regions display long, strip-shaped anomalies. This unusual reflection suggests the presence of underground structures or material changes, particularly underground cavities or voids.\u003c/p\u003e"},{"header":"5. Discussion and Analysis","content":"\u003cp\u003eBased on the analysis of GPR results at different frequencies, Figs.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e and \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e present the anomaly distribution map and cross-sectional diagram of the detection area, respectively. By combining the engineering geological data and measured results, the following conclusions can be drawn: At stake numbers 7\u0026thinsp;+\u0026thinsp;150 m to 7\u0026thinsp;+\u0026thinsp;170 m, within a depth of 0 to 2 m, there are loose soil hazards; at stake numbers 6\u0026thinsp;+\u0026thinsp;960 m to 6\u0026thinsp;+\u0026thinsp;970 m, within a depth of 0 to 5 m, and at stake numbers 6\u0026thinsp;+\u0026thinsp;870 m to 6\u0026thinsp;+\u0026thinsp;890 m, within a depth of 2 to 9 m, there are void hazards.\u003c/p\u003e\u003cp\u003eThe void anomaly area is associated with the large cranes at the port. Over time, the frequent use of these cranes has caused settlement in localized areas of the concrete, particularly where the crane loads are concentrated on both sides of the road. This has led to rapid settlement of the roadbed on either side. Due to the uneven nature of the settlement process, a void or cavity may have formed in the central part of the crane operating area, where the gap beneath the concrete was neither filled nor supported. This has resulted in the abnormal reflection waves detected by the GPR.\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eBased on the characteristics of the dike project, GPR detection and comprehensive analysis have been utilized to identify the types of hazards, their burial depths, and extents. Loose soil hazards were detected in the horizontal section between pile numbers 7\u0026thinsp;+\u0026thinsp;150 m and 7\u0026thinsp;+\u0026thinsp;170 m, at depths ranging from 0 to 2 meters. Void hazards were found in the horizontal sections between pile numbers 6\u0026thinsp;+\u0026thinsp;960 m and 6\u0026thinsp;+\u0026thinsp;970 m, at depths of 0 to 5 meters, and between pile numbers 6\u0026thinsp;+\u0026thinsp;870 m and 6\u0026thinsp;+\u0026thinsp;890 m, at depths of 2 to 9 meters. The detection results align closely with the adverse geological phenomena identified during the on-site investigation and the stratigraphic features revealed by drilling. These findings indicate that the dike is affected by loose soil and void hazards. Specifically, on the river-facing side, loose dike material with a higher moisture content was found, while near the crane area, voids and potential collapse hazards were detected. These results provide crucial data for the development of the subsequent remediation plan.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003ch2\u003eCompeting interests\u003c/h2\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConceptualization, R.H.; methodology, F.J.; software, L.G.; validation, J.Y.; formal analysis, J.N.; investigation, X.X.; resources, R.H.; data curation, F.J.; writing\u0026mdash;original draft preparation, R.H.; writing\u0026mdash;review and editing, R.H.; visualization, F.J.; supervision, X.X.; project administration, F.J.; funding acquisition, F.J. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgments\u003c/h2\u003e\u003cp\u003eThis work was supported by the National Key Research and Development Program of China (Grant No. 2021YFC3000103), the National Natural Science Foundation of China (Grant No. 41504081).\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data presented in this study are available on request from the corresponding author. The data are not publicly available due to institutional policies.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eLiu, X.\u003cem\u003e et al.\u003c/em\u003e Accurate gain method for ground-penetrating radar signals based on stationary wavelet packet transform. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e228\u003c/strong\u003e, 105473, doi:https://doi.org/10.1016/j.jappgeo.2024.105473 (2024).\u003c/li\u003e\n\u003cli\u003eEskandari Torbaghan, M.\u003cem\u003e et al.\u003c/em\u003e Automated detection of cracks in roads using ground penetrating radar. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e179\u003c/strong\u003e, 104118, doi:https://doi.org/10.1016/j.jappgeo.2020.104118 (2020).\u003c/li\u003e\n\u003cli\u003eZhang, Y.\u003cem\u003e et al.\u003c/em\u003e Confidence level-based size estimation of internal crack using multi-trace ground penetrating radar. \u003cem\u003eConstruction and Building Materials.\u003c/em\u003e \u003cstrong\u003e474\u003c/strong\u003e, 141124-141124，doi:https://doi.org/10.1016/j.conbuildmat.2025.141124 (2025).\u003c/li\u003e\n\u003cli\u003eHunziker, J., Meles, G. \u0026amp; Linde, N. Crosshole ground-penetrating radar full-waveform inversion by combining optimal-transport and least-squares distances. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e237\u003c/strong\u003e, 105655, doi:https://doi.org/10.1016/j.jappgeo.2025.105655 (2025).\u003c/li\u003e\n\u003cli\u003eDa, H. Deep Learning\u0026ndash;Based Framework for Bridge Deck Condition Assessment Using Ground Penetrating Radar. \u003cem\u003eJournal of Structural Design and Construction Practice\u003c/em\u003e,doi: https://doi.org/10.1061/JSDCCC.SCENG-1686 \u003cstrong\u003e30\u003c/strong\u003e (2025).\u003c/li\u003e\n\u003cli\u003eFeng, D., Fang, Z., Wang, X. \u0026amp; Ding, S. An efficient footprint-guided finite domain algorithm for common offset ground penetrating radar forward modeling. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e233\u003c/strong\u003e, 105617, doi:https://doi.org/10.1016/j.jappgeo.2024.105617 (2025).\u003c/li\u003e\n\u003cli\u003eRuan, W., Liu, H., Dong, H. \u0026amp; Liu, B. Enhanced Ground-Penetrating radar methods for horizon and dielectric characterization of Luvisols. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e,doi: https://doi.org/10.1016/j.compag.2025.110296 \u003cstrong\u003e234\u003c/strong\u003e, 110296-110296 (2025).\u003c/li\u003e\n\u003cli\u003eZatar, W., Nguyen, T. T. \u0026amp; Nguyen, H. Environmental effects on condition assessments of concrete structures with ground penetrating radar. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e203\u003c/strong\u003e, 104713, doi:https://doi.org/10.1016/j.jappgeo.2022.104713 (2022).\u003c/li\u003e\n\u003cli\u003eDahunsi, J., Pathirana, S., Cheema, M., Krishnapillai, M. \u0026amp; Galagedara, L. Estimating soil hydraulic conductivity from time-lapse ground-penetrating radar data in podzolic soils using the green-ampt model. \u003cem\u003eJournal of Hydrology\u003c/em\u003e,doi: https://doi.org/10.2139/ssrn.4939048 \u003cstrong\u003e657\u003c/strong\u003e, 133059-133059 (2025).\u003c/li\u003e\n\u003cli\u003eVasques, G. M.\u003cem\u003e et al.\u003c/em\u003e Ground penetrating radar (GPR) models of the regolith and water reservoir of an underground dam in the Brazilian semiarid region. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e206\u003c/strong\u003e, 104797, doi:https://doi.org/10.1016/j.jappgeo.2022.104797 (2022).\u003c/li\u003e\n\u003cli\u003ePashoutani, S. \u0026amp; Zhu, J. Ground Penetrating Radar Data Processing for Concrete Bridge Deck Evaluation. \u003cem\u003eJournal of Bridge Engineering\u003c/em\u003e,doi: .http://dx.doi.org/10.1061/(asce)be.1943-5592.0001566 \u003cstrong\u003e25,\u003c/strong\u003e (2020).\u003c/li\u003e\n\u003cli\u003eLi, K., Xie, X., Chen, H., Zhou, B. \u0026amp; Huang, C. Ground penetrating radar detection of steel fiber reinforced composite linings in shield tunnels: Experimental and field studies. \u003cem\u003eFrontiers of Structural and Civil Engineering\u003c/em\u003e,doi: https://doi.org/10.1007/s11709-025-1165-8 \u003cem\u003e \u003c/em\u003e1-15 (2025).\u003c/li\u003e\n\u003cli\u003eBigman, D. P. \u0026amp; Day, D. J. Ground penetrating radar inspection of a large concrete spillway: A case-study using SFCW GPR at a hydroelectric dam. \u003cem\u003eCase Studies in Construction Materials\u003c/em\u003e,doi: .http://dx.doi.org/10.1016/j.cscm.2022.e00975 \u003cstrong\u003e16\u003c/strong\u003e, (2022).\u003c/li\u003e\n\u003cli\u003eAlarifi, S. S., Ibrahim, E. \u0026amp; Kahtany, K. A. Ground-penetrating radar imaging of subsurface karst features controlling water leakage across Wadi Namar dam, south Riyadh, Saudi Arabia. \u003cem\u003eOpen Geosciences\u003c/em\u003e,doi: https://doi.org/10.1515/geo-2025-0781 \u003cstrong\u003e17\u003c/strong\u003e, 20250781-20250781 (2025).\u003c/li\u003e\n\u003cli\u003eBakker, M. A. J., Maljers, D. \u0026amp; Weerts, H. J. T. Ground-penetrating radar profiling on embanked floodplains. \u003cem\u003eNetherlands Journal of Geosciences-Geologie En Mijnbouw\u003c/em\u003e,doi: .http://dx.doi.org/10.1017/s0016774600021314 \u003cstrong\u003e86\u003c/strong\u003e, 55-61, (2007).\u003c/li\u003e\n\u003cli\u003eLee, S., Lee, J. H., Joh, C. \u0026amp; Cho, K. Improved delay-and-sum algorithm for ground-penetrating radar imaging considering refraction in multi-layered media. \u003cem\u003eStructures\u003c/em\u003e,doi: https://doi.org/10.2139/ssrn.4910087 \u003cstrong\u003e75\u003c/strong\u003e, 108700-108700 (2025).\u003c/li\u003e\n\u003cli\u003eGernant, C.\u003cem\u003e et al.\u003c/em\u003e Insights into the sea-level history of the South Shetland Islands from ground penetrating radar on Livingston Island, Antarctica. \u003cem\u003eQuaternary Science Reviews\u003c/em\u003e,doi: https://doi.org/10.1130/abs/2020am-356922 \u003cstrong\u003e359\u003c/strong\u003e, 109363-109363 (2025).\u003c/li\u003e\n\u003cli\u003eAksoy, A., Yigit, E., Caglayan, A. B. \u0026amp; Ozturk, M. Interpretation and analysis of scattering in steel fiber alignment in concrete: The H alpha decomposition method using fully polarized B-scan ground penetrating radar (GPR) image data. \u003cem\u003eJournal of Building Engineering\u003c/em\u003e,doi: https://doi.org/10.1016/j.jobe.2025.112472 \u003cstrong\u003e105\u003c/strong\u003e, 112472-112472 (2025).\u003c/li\u003e\n\u003cli\u003eColeman, Z. W. \u0026amp; Schindler, A. K. Investigation of ground-penetrating radar defect detection capabilities, influence of moisture content, and optimal data collection orientation in condition assessments of concrete bridge decks. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e202\u003c/strong\u003e, 104655, doi:https://doi.org/10.1016/j.jappgeo.2022.104655 (2022).\u003c/li\u003e\n\u003cli\u003eHe, W., Hao, T., Ke, H., Zheng, W. \u0026amp; Lin, K. Joint time-frequency analysis of ground penetrating radar data based on variational mode decomposition. \u003cem\u003eJournal of Applied Geophysics\u003c/em\u003e \u003cstrong\u003e181\u003c/strong\u003e, 104146, doi:https://doi.org/10.1016/j.jappgeo.2020.104146 (2020).\u003c/li\u003e\n\u003cli\u003eAyachi, R.\u003cem\u003e et al.\u003c/em\u003e Assessing YOLO models for real-time object detection in urban environments for advanced driver-assistance systems (ADAS). \u003cem\u003eAlexandria Engineering Journal\u003c/em\u003e \u003cstrong\u003e123\u003c/strong\u003e, 530-549, doi:https://doi.org/10.1016/j.aej.2025.03.077 (2025).\u003c/li\u003e\n\u003cli\u003eYeom, J.\u003cem\u003e et al.\u003c/em\u003e Automatic Detection of Borehole Breakout for Image Logs using YOLO algorithm. \u003cem\u003eGeoenergy Science and Engineering\u003c/em\u003e, 213925, doi:https://doi.org/10.1016/j.geoen.2025.213925 (2025).\u003c/li\u003e\n\u003cli\u003eAkdoğan, C., \u0026Ouml;zer, T. \u0026amp; Oğuz, Y. PP-YOLO: Deep learning based detection model to detect apple and cherry trees in orchard based on Histogram and Wavelet preprocessing techniques. \u003cem\u003eComputers and Electronics in Agriculture\u003c/em\u003e \u003cstrong\u003e232\u003c/strong\u003e, 110052, doi:https://doi.org/10.1016/j.compag.2025.110052 (2025).\u003c/li\u003e\n\u003cli\u003eHu, W.\u003cem\u003e et al.\u003c/em\u003e Robust crack detection in complex slab track scenarios using STC-YOLO and synthetic data with highly simulated modeling. \u003cem\u003eAutomation in Construction\u003c/em\u003e \u003cstrong\u003e175\u003c/strong\u003e, 106219, doi:https://doi.org/10.1016/j.autcon.2025.106219 (2025).\u003c/li\u003e\n\u003cli\u003eCai, Z., Zhang, Y., Li, J., Zhang, J. \u0026amp; Li, X. Synchronous detection of internal and external defects of citrus by structured-illumination reflectance imaging coupling with improved YOLO v7. \u003cem\u003ePostharvest Biology and Technology\u003c/em\u003e \u003cstrong\u003e227\u003c/strong\u003e, 113576, doi:https://doi.org/10.1016/j.postharvbio.2025.113576 (2025).\u003c/li\u003e\n\u003cli\u003eSong, B.\u003cem\u003e et al.\u003c/em\u003e YOLO-ELWNet: A lightweight object detection network. \u003cem\u003eNeurocomputing\u003c/em\u003e \u003cstrong\u003e636\u003c/strong\u003e, 129904, doi:https://doi.org/10.1016/j.neucom.2025.129904 (2025).\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e Wang H Q,.\u003cem\u003eet al.\u003c/em\u003e\u003c/strong\u003eImproved YOLOv5 for common underground pipeline detection using ground-penetrating radar. \u003cem\u003eChinese Journal of Geophysics\u003c/em\u003e. 2024, 67: 3588-3604.\u003c/li\u003e\n\u003cli\u003e.28 \u003cstrong\u003eYang B S.\u003cem\u003e et al.\u003c/em\u003e\u003c/strong\u003e Real-time detection method for underground targets using vehicle-mounted ground-penetrating radar. \u003cem\u003eActa Geodaetica et Cartographica Sinica\u003c/em\u003e. 2020, 49: 874-882.\u003c/li\u003e\n\u003cli\u003eZhi, Q.\u003cem\u003e et al.\u003c/em\u003e Application of an Improved YOLOv5 Algorithm in Real-Time Detection of Foreign Objects by Ground Penetrating Radar.\u003cem\u003e Remote Sensing. \u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 1895-1895 (2022).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Dike, YOLOv11, Ground Penetrating Radar, Loose, Void","lastPublishedDoi":"10.21203/rs.3.rs-6633296/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6633296/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn maintaining safe navigation along the Wenzaobang River, the dike structure is subject to wave erosion, which can lead to hazards such as surface subsidence, void formation, tilting of the flood control wall, and localized cracking. This study aims to investigate the distribution of potential hazards within the dike by constructing a geological model of loose and voided soil, based on regional geological conditions, and performing forward modeling simulations. The results reveal that loose soil within the dike generates chaotic waveforms in ground-penetrating radar (GPR) images, while voids manifest as arc-shaped reflections. Using the forward modeling outcomes and the YOLOv11 network structure for recognition, the analysis and interpretation of the GPR data indicate the following potential hazard locations: at station 7\u0026thinsp;+\u0026thinsp;150 m to 7\u0026thinsp;+\u0026thinsp;170 m (depth 0\u0026ndash;2 m), loose soil; at station 6\u0026thinsp;+\u0026thinsp;960 m to 6\u0026thinsp;+\u0026thinsp;970 m (depth 0\u0026ndash;5 m), and at station 6\u0026thinsp;+\u0026thinsp;870 m to 6\u0026thinsp;+\u0026thinsp;890 m (depth 2\u0026ndash;9 m), void formation. This research validates the integration of GPR and numerical simulation for geological hazard detection, demonstrating its applicability in complex environments and providing critical data for the design of dike reinforcement.\u003c/p\u003e","manuscriptTitle":"The Application of Ground Penetrating Radar in the Detection of Hidden Hazards in the Wenzaobang River Dike in Shanghai","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-25 11:36:11","doi":"10.21203/rs.3.rs-6633296/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"34d66e38-6e7c-4de7-86d6-910c8c937eee","owner":[],"postedDate":"July 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":50663191,"name":"Earth and environmental sciences/Natural hazards"},{"id":50663192,"name":"Physical sciences/Engineering"}],"tags":[],"updatedAt":"2026-01-22T14:56:18+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-25 11:36:11","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6633296","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6633296","identity":"rs-6633296","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}