Vibrations are ubiquitous, and vibration control technologies continue to evolve alongside scientific and technological advancements across various fields. In the fields of civil engineering, transportation and environment engineering, the rapid development of rail transit has exacerbated the adverse effects of ambient vibrations. To satisfy demands for comfortable living environments and, more importantly, the stringent micro-vibration requirements for operating precision instruments and manufacturing high-tech products, various vibration control measures are being actively developed. Three primary vibration control strategies are commonly employed in civil engineering: attenuating the excitation at the source, blocking the vibration transmission path, and directly protecting the target structure. Focusing on pile barriers and open (or in-filled) trench barriers frequently used for path control, as well as the surrounding soil in which these barriers are embedded, this paper presents four main contributions. First, two classification schemes are introduced: wave barriers are categorized into traditional and periodic wave barriers based on their underlying design theories; and the embedding soil is modeled with increasing complexity as single-phase, saturated, or unsaturated soil. Second, a brief review of the research progress on traditional wave barriers in single-phase, saturated, and unsaturated soils is provided. Third, based on the author’s nearly 20 years of research experience with periodic wave barriers, a comprehensive and systematic review of their development in single-phase, saturated, and unsaturated soils is presented. Furthermore, there is an urgent need for research on periodic wave barriers under moving load excitations. Finally, a summary and future outlook are provided. By focusing on the advancement of wave barriers in civil engineering, particularly periodic wave barriers, this paper serves as a valuable reference for researchers engaged in ambient vibration control.

In the semi-logarithmic plane of void ratio and stress, the isotropic compression line of reconstituted soil typically approaches the Normal Compression Line (NCL) from below, whereas the isotropic compression line of structured soil first crosses to the right of the NCL and subsequently approaches it from above. Consequently, structured soil exhibits a high modulus during the initial stage of compression, while structural collapse in the intermediate and late stages results in a reduced modulus and significant deformation. To simulate these one-dimensional compression characteristics, an exponential evolution law is assigned to the Moving Normal Compression Line (MNCL) within the theoretical framework of the Unified Hardening (UH) model, thereby introducing the Virgin Compression Line (VCL) for structured soil. Analogous to the NCL of reconstituted soil, the VCL of structured soil represents the compression curve of a structured soil that has not experienced any prior stress history during its primary compression. Using the VCL as a reference, an appropriate volumetric deformation evolution equation is formulated to establish a one-dimensional constitutive description for structured soils. The results demonstrate that the proposed constitutive description is continuous and smooth, with the compression line of structured soil consistently remaining below the VCL. Comparisons between the model predictions and experimental data indicate that the established one-dimensional constitutive model accurately captures the initial high modulus, the intermediate low modulus, and the ultimate convergence of the compression curve toward the NCL of reconstituted soil during both isotropic and confined compression. This study provides a theoretical reference for the engineering simulation and computational analysis of structured soil foundations.

To address the unsaturated seepage problem in surface soil layers under complex boundary conditions, linearized theoretical models for unsaturated seepage during two distinct infiltration stages, pre-ponding and post-ponding, are established using mathematical techniques such as dimensionless linearization and the principle of equation homogenization. Analytical expressions for hydraulic conductivity, matric suction head, and volumetric water content during the corresponding unsaturated seepage processes are derived. Additionally, a method for calculating surface ponding time is proposed, and the reliability of this analytical approach is verified numerically. The results indicate that under periodic surface boundary conditions, both the volumetric water content and matric suction head at characteristic points exhibit periodic variations consistent with the surface boundary. Compared to surface conditions, characteristic points near the surface display a distinct phase lag. Furthermore, the peaks and troughs of the periodic evolution curves for volumetric water content and matric suction head increase over time. The period of the seepage rate evolution curve matches that of the surface boundary, but its amplitude is significantly reduced, indicating that the soil layer exerts an attenuation or "peak-clipping" effect on the periodic fluctuations of the seepage rate, an effect that becomes more pronounced with increasing depth.Under continuous infiltration conditions, prior to surface ponding, the upper portion of the soil layer is in an infiltration state, whereas the lower portion is in an evaporation state due to the influence of the groundwater table. Following surface ponding, the entire soil layer transitions to a state of sustained infiltration, ultimately approaching saturation. The evolution characteristics of volumetric water content and matric suction head within the soil layer under continuous infiltration reflect the combined effects of the surface boundary and groundwater. This study offers significant theoretical and practical value for analyzing unsaturated seepage hydrological processes and for the prevention and mitigation of geological hazards.

To address the unclear long-term coupling mechanisms between vehicle loads and environmental thermal loads in the dynamic response analysis of pavement structures, this study proposes an analytical model that comprehensively considers the coupled effects of moving vehicle loads and diurnal temperature variations. By utilizing thermoelastic theory to describe the pavement surface, base, and subgrade, and introducing the theory of a thermally coupled saturated porous elastic medium to simulate the soft soil foundation, a fully coupled analytical model encompassing the vehicle load, thermal load, and layered pavement system is established. Based on this model, combined with the Laplace-Hankel integral transforms, the system’s solutions in the transform domain are derived, and solutions in the physical time domain are obtained through numerical inversion. The results indicate the following: Under diurnal temperature variations, thermal fluctuations attenuate rapidly with increasing depth, with daily variations primarily concentrated within a shallow depth of 0.3 m below the pavement surface; These shallow temperature changes directly induce thermal deformation in the surface layer, which further propagates downward to trigger dynamic responses in the subgrade and foundation, and the excess pore water pressure in the soft soil foundation peaks at a depth of approximately 2 m; Under the coupled action of thermal and vehicle loads, the maximum vertical displacement at the pavement surface and the variation in pore water pressure within the soft soil foundation reach 61% and 240% of the values induced by the vehicle load alone, respectively. This research provides a theoretical framework and quantitative reference for accurately evaluating the dynamic responses and long-term performance of pavement structures under the coupled effects of complex environmental and traffic loads.

To address the limitation that existing research on the lateral deformation capacity of precast bridge piers predominantly focuses on single connection types, rendering it unsuitable for hybrid-connected precast piers with distinct structural configurations and load transfer mechanisms, a calculation method for the lateral deformation capacity of such bridge piers is proposed. This study investigates a precast bridge pier utilizing hybrid connections, specifically, unbonded prestressing tendons, grouted sleeves, and shallow socket connections, employed in the Beijing-Xiong’an Transit Express Line. Based on a full-scale quasi-static test, a refined numerical model is established using ABAQUS, incorporating concrete damage, bond-slip behavior of longitudinal reinforcement, and the mechanical characteristics of unbonded prestressing tendons. Through parametric sensitivity analysis, the influence of various design parameters on the Plastic Hinge Length (PHL) of the precast bridge pier is elucidated. Furthermore, an empirical formula for calculating the equivalent PHL applicable to such piers is proposed via nonlinear regression analysis. Building upon the equivalent plastic hinge theory, a method for calculating the lateral displacement at the pier top is proposed, accounting for both the flexural deformation of the pier body and the rotational deformation of the socketed section. The results indicate that the PHL increases with greater superstructure dead load, shear span ratio, longitudinal reinforcement ratio and yield strength, initial prestressing force, and socket depth. Conversely, it decreases as concrete strength and sleeve length increase. The equivalent PHLs and the load-displacement curves calculated using the proposed method align well with the experimental results, exhibiting an ultimate displacement prediction error of merely 3.96%. The results provide a theoretical reference for the deformation capacity evaluation and seismic design of precast bridge piers with hybrid connections.

To address the critical issues of mechanical degradation in the ballast bed at girder ends and the resulting deterioration of track regularity caused by the thermal expansion and contraction of long-span bridges, a bridge-ballast-sleeper coupling model is established based on the coupled discrete element method (DEM) and multi-body dynamics. This study reveals the meso-mechanical behaviors of ballast particles at girder ends, specifically their motion characteristics, compactness evolution, and force chain transmission mechanisms, under the effects of main girder deformation and sleeper group displacement. The results indicate that during the expansion and contraction of the main girder, the ballast particles at the girder end and the bottom of the ballast bed exhibit the highest movement velocities, while those between the sleepers show the lowest velocities with a significant upward arching tendency. The lateral displacement of ballast particles remains consistently at approximately 2 mm. The longitudinal displacement of the particles beneath the sleeper is significantly greater than that at the sleeper ends, increasing with depth and reaching a peak of 7.36 mm. The compactness of the ballast bed beneath the sleepers at the girder end decreases as the main girder elongates and increases as it contracts, whereas the compactness in the sleeper box regions. The contact force between ballast particles increases during the elongation of the main girder, and initially decreases before increasing during contraction. The internal force chains within the ballast bed are transmitted from the bottom to the surface at an angle of approximately 45°, with an average peak contact force of 18.90 N. Under the displacement of the sleeper group, the ballast on the sides of the sleepers undergoes longitudinal flow, while the ballast beneath the sleepers experiences rotational dislocation. The compactness in the sleeper bottom and crib regions fluctuates, and the overall compactness of the ballast bed can decrease by 1.89%. Contact forces are transmitted downward from the sleeper bottom and sides to the base of the ballast bed at an angle of roughly 45°. The contact forces on the sleepers are primarily concentrated on the sides and bottom, resulting in a longitudinal resistance of approximately 10 kN per sleeper, which is lower than the longitudinal resistance measured in single-sleeper tests. These findings clarify the mechanisms by which bridge expansion deformation influences the meso-mechanical behavior of the ballast bed at girder ends, providing theoretical support for understanding the evolution of track regularity at girder ends and ensuring the safety of railway operations.

To address the difficulty of cross-validating multi-source data for in-service road-rail sea-crossing cable-stayed bridges, a dual-reference closed-loop evaluation method is proposed. This method utilizes a calibrated Finite Element Model (FEM) as the theoretical baseline, and historical Structural Health Monitoring (SHM) records along with routine inspection results as the temporal baseline. Under a unified relative difference index, a closed-loop evidence chain is constructed, encompassing the overall stiffness, dynamic characteristics, and mechanical states of key components. First, static and dynamic vehicle load tests are conducted to obtain the displacement, strain, and influence lines of the main girder, which are then compared with theoretical results. Second, Operational Modal Analysis (OMA) is employed to extract the first eight modal parameters, which are subsequently verified against theoretical values and historical data. Third, vibration acceleration signals from representative stay cables are collected; the cable forces are calculated based on the modified taut-string theory and validated against historical records and FEM-calculated distributions. Finally, preprocessing procedures, including environmental baseline correction, detrending, and band-pass filtering, are applied to mitigate interference from non-structural factors, thereby establishing a verifiable and traceable multi-source closed-loop evaluation process. The results indicate that the relative deviation between the measured static-load displacement at the mid-span and the FEM-calculated value is less than 6.5%, and the dynamic load influence line aligns closely with the theoretical predictions, demonstrating that the bridge’s overall stiffness and load transfer characteristics remain stable. The frequencies of the first eight modes are generally consistent with the theoretical and historical baselines, with the relative difference rates for most modes not exceeding 3%, indicating no significant degradation in overall dynamic stiffness. The relative deviation in the forces of representative stay cables is within 6.5%, suggesting that the mechanical states of key components are fundamentally stable. This method achieves the multi-dimensional cross-validation of vehicle load tests, operational modal identification, and cable force measurements, providing a reliable reference for routine structural inspections and operation and maintenance decision-making for in-service road-rail sea-crossing cable-stayed bridges.

To address the challenges of numerous construction procedures, complex stress conditions, and difficulty in controlling surrounding rock deformation at the intersection of the No. 2 cross-tunnel and the main tunnel of the Qunke Tunnel on the Xi’an-Chengdu Railway, a roof-breaking excavation scheme and associated mechanical analysis are investigated. First, a comprehensive roof-breaking excavation and support scheme consisting of “cross-tunnel climbing + reinforcing ring support + temporary-to-permanent support + symmetrical excavation and support” is proposed. Subsequently, a numerical model is established using Midas finite element software to finely analyze the dynamic mechanical response characteristics and three-dimensional spatial effects of the support structure and surrounding rock during the roof-breaking construction process. Finally, combined with field real-time monitoring data, the analysis focuses on the stress variation characteristics of the support structure and surrounding rock at the tunnel intersection, as well as crown settlement and lateral convergence deformation patterns around the tunnel periphery. The results show that during construction, the stress of the primary support structure remains within the range of 0.15 MPa to 0.76 MPa, and the maximum axial force in the steel arch frame is 343 kN. This support system effectively reduces the stress and settlement deformation of the primary support induced by surrounding rock pressure. Construction of the intersection induces stress redistribution in the surrounding rock near the reinforcing ring, making this area prone to stress concentration. The maximum crown settlement at Sections 1-3 is 1.12 mm, and the maximum lateral displacement at the arch springline is 1.06 mm; both values are relatively small, indicating that the intersection remains in a stable and controllable state overall. During the excavation and support process of the left-side upper bench at the intersection, the left and right arch springlines and the right arch shoulder transition from compression to tension. These locations should be closely monitored during construction, and timely support should be provided to prevent local rockfalls, ensuring that the entire construction process remains safe and stable. The research findings can provide important references for roof-breaking construction, support structure design, and surrounding rock stability control at tunnel cross-tunnel and main tunnel intersections under similar complex conditions.

To address the issue where piston wind generated by train movement during subway operations significantly alters the flow field distribution inside jet fans, thereby causing fluctuations in ventilation efficiency and reduced operational stability, an integrated numerical model coupling the linear motion of the train with the rotational motion of the fan blades is proposed and validated through field experiments. This study determines the dynamic variation laws of the flow mechanisms and performance parameters within the jet fan under the influence of piston wind, elucidating the impact mechanisms of factors such as piston wind morphology and train speed on fan performance. The results indicate that the variation patterns of the flow field characteristics inside the jet fan are closely related to three distinct forms of piston wind. The superposition effect of the velocity field and the airflow field is significant, demonstrating clear instantaneous and corresponding variation characteristics. As train speed increases, the suppressive effect of the piston wind on both the static pressure difference between the fan inlet and outlet and the exhaust air volume gradually intensifies. The reduction amplitude of the static pressure difference exhibits a positive linear relationship with the train speed, whereas the suppression amplitude of the exhaust air volume shows a positive linear relation-ship with the square of the train speed. In a horseshoe-shaped tunnel with a blockage ratio of 0.3, a train speed of 75 km/h serves as the critical threshold. At this speed, the stable operation of the jet fan is maintained, and the reduction in ventilation efficiency is minimized. These findings provide a theoretical reference for improving the working efficiency, optimizing operational parameters, and evaluating the performance of jet fans in subway tunnels.

To address control issues such as excessive longitudinal impulses and inflexible formations in long heavy-haul trains under traditional pneumatic braking system architectures, this paper proposes a Tube Model Predictive Control (Tube-MPC) method based on virtual coupling technology and robust constraints. This method adopts a short-consist operation mode, dividing long heavy-haul trains into multiple independent train units according to operational requirements to enhance operational efficiency and flexibility while reducing longitudinal impulses within individual trains. First, a multi-body dynamic model of the Heavy-Haul Virtually Coupled Train Set (HHVCTS) is constructed by comprehensively considering factors such as electric braking force, pneumatic braking force, coupler force, and running resistance, followed by model discretization. Second, based on this dynamic model, a multi-objective cost function is formulated to balance speed tracking accuracy, coupler force suppression, energy efficiency, and system stability, while incorporating constraints such as speed limits, coupler force safety limits, and actuator physical characteristics into the overall optimization framework. Third, the Tube-MPC method is employed to design a cooperative control strategy for the locomotives of each train unit within the HHVCTS. By designing terminal controllers and terminal constraint sets, sufficient conditions for controller stability are derived, demonstrating the feasibility and stability of the proposed control algorithm. Finally, simulation verification is conducted under two typical operating scenarios: gentle gradients and long downhill grades. The results indicate that the proposed cooperative control strategy achieves stable spacing regulation among train units. The average maximum speed error and spacing error of the system converge to 0.44 m/s and 0.66 m, respectively, realizing high-precision tracking of the target speed profile. During acceleration, deceleration, and on extended steep downgrades, the peak coupler force reaches approximately 330.4 kN—merely 22% of the safety limit (1500 kN). Quantitative results verify that the proposed method effectively ensures the robust stability of the system while satisfying stringent tracking error constraints, providing a theoretical basis and algorithmic foundation for the engineering deployment of heavy-haul virtual coupling technology.

To fully extract the time-frequency domain features from the power curves of S700K switch machines and to address the challenges of fault diagnosis under small-sample and imbalanced conditions, a fault diagnosis method combining the Short-Time Fourier Transform (STFT), a two-dimensional Convolutional Neural Network (2DCNN), and an improved Extreme Random Forest (ERF) is proposed. First, the collected one-dimensional power curves of the switch machine are converted into two-dimensional time-frequency images with dimensions of 16×16, 28×28, and 32×32 using STFT, achieving a joint representation of time-domain and frequency-domain information. Second, an improved 2DCNN architecture is constructed to extract multi-scale deep features from the time-frequency images. Third, the three sizes of time-frequency images are respectively input into the 2DCNN model for comparison to determine the optimal image size, and feature indicators are extracted through multiple convolutional, pooling, and fully connected layers. Finally, to enhance classification accuracy and the global generalization performance of the model, the Bayesian Optimization (BO) algorithm is employed to optimize the hyperparameters of the ERF, specifically the optimal number of trees and the optimal number of splits. Experimental results demonstrate that transforming one-dimensional data into two-dimensional time-frequency images enhances the discriminability of transient fault features. Furthermore, utilizing the 2DCNN to extract key features from the time-frequency images improves the hierarchy and robustness of the feature representation. By inputting the extracted features into the BO-ERF fault diagnosis model, an accuracy of 98.44% is achieved. Compared with other fault diagnosis models, the proposed method exhibits superior diagnostic performance and faster diagnostic speeds. Furthermore, utilizing the 2DCNN to extract key features from the time-frequency images improves the hierarchy and robustness of the feature representation. By inputting the extracted features into the BOERF fault diagnosis model, an accuracy of 98.44% is achieved. Compared with other fault diagnosis models, the proposed method exhibits superior diagnostic performance and faster diagnostic speeds.

Traditional single-source data-based fault diagnosis methods for railway switch machines are limited by insufficient fault feature representation from a single data modality. When dealing with complex fault types, these methods often suffer from low diagnostic accuracy and an inability to precisely and finely locate faults, making them inadequate for practical applications. To improve the diagnostic performance of the ZYJ7 electro-hydraulic switch machine, this paper proposes a deep learning-based fault diagnosis method based on feature-level multi-source data fusion using real-world oil pressure, current, and power data. First, multi-modal data from oil pressure, current, and power signals are normalized, denoised, and subjected to spatiotemporal feature extraction. Then, bilinear interpolation is employed to unify the dimensions of feature matrices across different modalities, and an attention mechanism is introduced to achieve dynamic weighted fusion of these feature matrices. Finally, the fused features are fed into a hybrid neural network composed of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for classification. The proposed method is compared with single-source data models, simple voting methods, simple concatenation methods, and fixed-weight fusion models. Experimental results show that the proposed feature-level multi-source data fusion model improves diagnostic accuracy by 7.19%, 4.38%, 3.13%, and 1.88%, respectively, demonstrating its effectiveness and providing a valuable reference for fault diagnosis of railway switch machines.

To address the need for defect identification in precision components of electric multiple unit (EMU) bogies during operational image inspection, this paper proposes a recognition algorithm for detecting loose and missing bogie bolts by combining template matching with an improved YOLOv5 model. The proposed algorithm first performs bogie extraction. During bogie extraction, bilinear interpolation is utilized to downscale the images, significantly reducing computational overhead. Simultaneously, normalized cross-correlation template matching is employed to maintain high-precision target localization on the downscaled images. Then, the proposed algorithm performs defect recognition, for which an improved YOLOv5 model is applied. For the defect recognition task, the model improvements mainly include three aspects: first, to address the large parameter size and high computational cost of the baseline model, a CNN-based Cross-Scale Feature-Fusion Module (CCFM) replaces the Path Aggregation Network (PANet), effectively reducing both the parameter count and overall model size; second, to meet the requirements for detecting tiny targets within the dataset, a small-object detection layer is added to enhance the capture of minute defects; third, to overcome the limited feature perception range, Large Selective Kernel (LSK) modules are embedded into the shallow network layers to expand the receptive field and improve detection accuracy. Finally, the bogie extraction efficiency and defect recognition performance of the algorithm are experimentally verified. Experimental results demonstrate that, in the bogie extraction stage, the average processing time of template matching on downscaled images is reduced by approximately 92%, from 929.58 ms to 74.87 ms per image. In the defect recognition stage, compared with the baseline model’s mean Average Precision (mAP) values of 89.1% at an IoU threshold of 0.5 (mAP@0.5) and 43.1% over IoU thresholds from 0.5 to 0.95 (mAP@0.5:0.95), respectively, the improved model achieves 91.5% and 46.2% while simultaneously reducing the parameter count by approximately 23%, yielding improvements of 2.4% and 3.1%. The proposed method effectively enhances both the efficiency and accuracy of defect detection for precision components in EMUs, providing a valuable technical reference for safe operation and maintenance.

To address the prevalent issues of uneven brightness distribution, loss of high-frequency details, and cross-domain feature mismatch in railway low-light images at night, a railway low-light image enhancement method based on a multi-level decomposition–fusion mechanism is proposed. First, a collaborative architecture integrating a multi-level residual decomposition module and a multi-scale feature fusion mechanism is constructed to achieve global illumination correction, local contrast enhancement, and high-frequency detail restoration. Second, a luminance-aware attention mechanism is introduced to establish a railway low-light image enhancement framework focused on brightness regulation and detail enhancement. By utilizing learnable weights, the enhancement regions are adaptively adjusted, effectively suppressing overexposure while enhancing details in dark areas. Third, the fusion mechanism enables the dynamic integration of cross-level features, optimizing the synergistic representation of global structures and local textures. Finally, the proposed model is compared with various baseline methods, and ablation studies are conducted to validate the effectiveness of each module. Experimental results indicate that the proposed method significantly improves railway image enhancement performance. On the LOLv1 dataset, it exhibits superior detail restoration and exposure control capabilities, with the Peak Signal-to-Noise Ratio (PSNR) improving by 0.63~9.13 dB and the Structural Similarity Index (SSIM) increasing by 0.013~0.325. Ablation experiments further confirm the contribution of the proposed modules to overall model performance. Furthermore, a dedicated railway low-light image dataset (RLL) is constructed. Through comparative and visual analyses against multiple existing low-light image enhancement methods, the superiority of the proposed method is robustly validated.

Under hazy weather conditions, traffic surveillance images often suffer from low contrast, blurred edges, and degraded distant visual information, which severely compromise the accuracy of road target detection and traffic event recognition. Existing image dehazing methods, which are primarily based on physical models or end-to-end network architectures, continue to exhibit limitations in structural detail preservation, edge continuity enhancement, and adaptability to complex traffic scenarios. To address these issues, this paper proposes an improved Vision Transformer (ViT)-based traffic image dehazing method that integrates the characteristics of progressive optimal filtering and guided filtering. First, an AO-Guided method is introduced at the input stage for image enhancement by leveraging the synergistic mechanism of progressive optimal extreme filtering and guided filtering, this approach enhances the edge and structural sensitivity of the images. Second, an AO-Guided feature enhancement module is embedded during the encoding stage to improve the network’s ability to model detailed information in mid- and long-range fog-degraded regions. Third, a joint loss function and an intermediate supervision mechanism are designed to guide the network in achieving the collaborative optimization of global structure preservation and local detail restoration. Finally, dehazing experiments are conducted on the RESIDE real-world hazy traffic image dataset. The results demonstrate that, compared to DehazeFormer, a representative ViT-based dehazing method, the proposed approach improves three key structural sensitivity metrics: the ratio of visible edges and the average gradient ratio increase by 13.1% and 40.5%, respectively, while the proportion of black pixels decreases by 25%. Consequently, the proposed method significantly enhances the dehazing performance of traffic images under hazy conditions.

Infrared imaging technology is not limited by lighting conditions and exhibits strong anti-interference capabilities, enabling stable operation at night and in complex, harsh environments. Therefore, it is a critical technology for vehicle-mounted intelligent systems to achieve all-weather environmental perception. However, infrared images generally suffer from blurred edges and a lack of detailed features, making it difficult to accurately detect small-scale targets such as pedestrians and failing to meet the real-time detection requirements of vehicle-mounted scenarios. To address the issues of feature attenuation and ambiguous localization in infrared images, a global context space object detection network named Global Context Space-You Only Look Once (GCS-YOLO) is proposed. First, a Global Adaptive Feature Extraction Module (GFEM) is designed to improve the backbone network. By introducing a Global Channel Attention (GCA) mechanism and adopting a residual structure with progressively selected kernels, the model adaptively modulates its receptive field to extract feature information at various scales. Second, a Multi-Spatial Attention Feature Pyramid Network (MSA-FPN) is designed. By using a coordinate attention module to enhance deep feature maps containing positional information and introducing lateral skip connections to incorporate shallow-layer information, the detection accuracy for small and weak targets is improved. Finally, the Minimum Point Distance Intersection over Union (MPDIoU) loss function is introduced. By minimizing the Euclidean distances between the top-left and bottom-right corners of the predicted and ground-truth bounding boxes, the accuracy of bounding box regression is enhanced. Experimental results demonstrate that, compared to YOLOv5, GCS-YOLO achieves an mAP@0.5 of 81.1% and an mAP@0.5:0.95 of 49.5% on an open-source infrared dataset, representing improvements of 10.4% and 7.4%, respectively. Furthermore, it operates at a processing speed of 26.4 FPS, successfully meeting the real-time detection requirements for vehicle-mounted applications. Compared to existing algorithms, GCS-YOLO demonstrates significant advantages in infrared object detection accuracy, providing effective technical support for the all-weather operation of intelligent driving systems.

Dual-mode communication technology, which integrates Power Line Communication (PLC) and micro-power wireless communication, has garnered significant attention for its ability to effectively mitigate communication interruptions caused by power line interference in smart distribution grids. To address the high deployment costs associated with micro-power wireless communication modules during the large-scale application of dual-mode communication technologies, this paper proposes an optimal deployment method for dual-mode communication nodes under constrained costs. First, a dual-tree network topology for smart distribution grids is constructed, and an optimal deployment scheme for dual-mode nodes is designed. Then, an optimization problem is formulated with the objective of minimizing the weighted sum of the signal attenuation rate and deployment costs. Finally, to solve this multi-constrained mixed-integer programming problem, it is decomposed into two scenarios for separate resolution: one with a fixed number of deployed dual-mode devices, the Deep Deterministic Policy Gradient (DDPG) algorithm is employed to optimize the deployment locations of the wireless communication modules. For the scenario with an unfixed optimal cost, a Dynamic Programming (DP) algorithm is adopted to optimize both the quantity and specific deployment locations of the dual-mode modules. Research results indicate that when the number of deployed dual-mode nodes is fixed, system performance degrades as the distance from ordinary and relay nodes to the concentrator increases. However, the DDPG-based deployment scheme consistently outperforms baseline strategies, including random deployment, prioritizing nodes with poor PLC performance, and prioritizing nodes with superior wireless performance. Furthermore, compared to the random deployment and relay node-first schemes, the DP-based scheme reduces the objective function value by 6.8% and 3.8%, respectively, demonstrating the effectiveness of the proposed approach in balancing deployment costs and system performance.

Two-Way Optic-fiber Time Transfer (TWOTT) is currently one of the most accurate time transfer methods in terms of uncertainty performance, capable of achieving repeatability on the order of hundreds of picoseconds. As a critical influencing factor, link calibration directly affects the accuracy of time transfer results. To address current scarcity of research on TWOTT link calibration, the lack of standardized methods, and the difficulties in verifying calibration results, this study proposes a TWOTT link calibration method based on Common Clock Difference (CCD). First, a time transfer link composed of two TWOTT devices is established. Calibration experiments are conducted under common-clock conditions, and the CCD method is employed to obtain the link calibration results. Then, an uncertainty evaluation is performed to determine the combined standard uncertainty of the link calibration. Finally, a cross-validation method utilizing experimental results from multiple time transfer links is proposed. Under laboratory conditions, different time transfer links are used to simultaneously compare two time scales, thereby verifying the TWOTT link calibration results. The experimental results demonstrate that the calibration value of the time transfer link composed of two TWOTT devices is 0.189 ns, with a combined standard uncertainty of 81 ps. When three different time transfer links are used simultaneously to compare two time scales in the laboratory, the differences among the obtained clock offsets all fall within the range of comparison uncertainty. These results verify the effectiveness and accuracy of the proposed TWOTT link calibration method, providing a technical reference for the calibration and verification of fiber-optic time transfer links.