Due to the rebars layer’s shielding effect on Ground Penetrating Radar (GPR) waves, the hyperbolic clutter generated by the rebars interferes with the echoes from void beneath them. The overlapping waveforms of both signals result in attenuation and distortion of the void signals, making it difficult to identify void defects under the rebar. This study proposes an unsupervised generative network model based on Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs). Through a shared latent space, mapping is achieved between two image domains, effectively eliminating the multiple reflection interference signals caused by the rebar while accurately reconstructing the void defects, generating GPR B-Scan images without rebar clutter. Additionally, the channel and spatial attention module (CSA) is implemented into the model to help the network to better focus on the essential information in GPR images. The proposed model was validated through ablation and comparative experiments using synthetic data. Finally, real GPR data from the Husa Tunnel were used to verify the model’s effectiveness in practical engineering applications. The results showed that this model is highly effective; it improves the visibility of void defects signals, thereby enhancing the interpretability of GPR data for tunnel lining inspections.

Under the influence of a variety of unfavorable factors, slope instability may occur, so the stability of the slope needs to be analyzed. In this paper, combined with the geological investigation data, the slope engineering geological profiles are derived based on 2D inversion maps of the high-density resistivity (HDR) method to obtain the overburden layer thickness. Then, a finite element model is established based on the mechanical and geometric parameters of the slope to analyze the displacement and deformation characteristics and obtain information about the potential sliding surface. The above analysis results are combined to determine the effective monitoring field and deploy (Fiber Bragg Grating) FBG strain detection piles for deep strain monitoring. The analysis showed that the shape and location of the potential sliding surface of the slope from the numerical simulation are consistent with those delineated by the HDR profile. The monitoring results show the rock and soil on the north side of the slope are soft and have poor stability, which is consistent with the HDR method detection results. Adopting the HDR method to identify the slope’s overall geological structure combined with FBG strain detection piles to obtain the deep deformation provided an effective monitoring technique for slope stability assessment.

Rebar constitutes a crucial element within tunnel lining structures, where its precise arrangement plays a pivotal role in determining both structural stability and load-bearing capacity. Due to the rebar’s high dielectric constant approaching infinity, radar signal reflections are intensified, manifesting as distinct hyperbolic patterns within radar imagery. By performing convolutional operations, these hyperbolic features of rebar can be effectively extracted from radar images. Building upon the feature extraction capabilities of the ResNet50 model, this study introduces a Deformable Attention to Capture Salient Information (DAS) mechanism, employing deformable and separable convolutions to enhance rebar localization and concentrate on hyperbolic features resulting from multiple reflections. Before the Region Proposal Network (RPN) and region of interest (ROI) pooling stages in Faster R-CNN, this study integrates a hyperbolic attention (HAT) module. Through refined distance metrics, the hyperbolic attention mechanism enhances the network’s Precision in identifying rebar hyperbolic features within feature maps. To ensure robustness across diverse conditions, this study utilizes a simulated public dataset for tunnel linings, alongside real data from the Husa Tunnel, to create a comprehensive ground-penetrating radar image dataset for tunnel linings. Experimental results indicate that the proposed model achieved an Average Precision of 94.93%, reflecting a 3.14% increase compared to the baseline model. Lastly, in a random selection of 50 radar images for testing, the model achieved a rebar detection accuracy of 93.46%, representing an enhancement of 0.94% over the baseline model.

The secondary lining of a tunnel is a critical load-bearing component, whose stability and structural integrity are essential for ensuring the overall safety of the tunnel. However, identifying lining structures and estimating their thickness using ground-penetrating radar (GPR) remain challenging due to several inherent limitations. First, the limited electromagnetic contrast between the primary and secondary linings results in weak interface reflections in GPR imagery, thereby hindering accurate delineation. Second, construction errors such as over-excavation or under-excavation often lead to complex interface geometries, further complicating the interpretation of GPR signals. To address these challenges, we propose an enhanced YOLOv8-seg network capable of performing pixel-level segmentation on GPR images to accurately delineate secondary lining regions and estimate their thickness. The model integrates a convolutional block attention module (CBAM) to refine feature extraction by emphasizing critical characteristics of the two interface layers through channel-wise and spatial attention mechanisms. The model is first pretrained on the COCO dataset and subsequently fine-tuned via transfer learning using a hybrid GPR dataset comprising real-world measurements and numerically simulated data based on forward modeling. Finally, the model is validated on real-world GPR measurements acquired from the Longhai tunnel. Experimental results demonstrate that the proposed method reliably identifies secondary tunnel linings and accurately estimates their average thickness.

Existing monocular video human motion recovery methods can already estimate global trajectories and local poses in world coordinates. However, their results are still largely driven by visual fitting and often suffer from foot penetration, hovering, and foot skating, as well as trunk lean and center-of-mass sway that are inconsistent with the underlying dynamic balance mechanism of continuous walking. Existing methods based on physics engines or explicit dynamics are usually complex to solve and difficult to integrate tightly with current reconstruction pipelines. Methods based on differentiable physical priors are easier to integrate, but most focus on contact consistency or static stability modeling, and therefore cannot fully characterize the dynamic stability of continuous gait, which is jointly determined by velocity and support transition. To address this issue, we propose a backend physical refinement method with dynamic stability constraints for monocular 3D gait reconstruction. Given an initial human motion estimate in world coordinates, the method constructs velocity-dependent dynamic stability constraints based on the extrapolated center of mass (XCoM) theory, and jointly refines the full sequence by incorporating the effective center of pressure, foot-ground contact constraints, kinematic anchoring, and temporal smoothness. Experiments show that the proposed method significantly mitigates penetration, hovering, and foot skating, while improving event-based lateral stability and XCoM trajectory consistency, thereby better preserving natural gait geometry while enhancing physical plausibility.