A multi-level integrated sensing system and method for monitoring and early warning of highway infrastructure.

Abstract

This invention discloses a multi-level integrated sensing system and method for monitoring and early warning of highway infrastructure, relating to the field of intelligent transportation, aiming to meet the needs of safety monitoring and intelligent early warning throughout the entire life cycle of highways. The system includes an end-side optimized deployment module, a fiber-optic-video dual-channel transmission module, a fiber-optic physical field inversion module, a fiber-optic-video redundant sensing module, and a multi-source fusion dynamic risk management module. The end-side optimized deployment module sends the collected fiber-optic sensor signals and video monitoring data to the fiber-optic-video dual-channel transmission module. The fiber-optic physical field inversion module converts the fiber-optic sensor signals processed by the fiber-optic-video dual-channel transmission module into physical parameter information and sends it to the fiber-optic-video redundant sensing module. Based on the physical parameter information and video monitoring data, the fiber-optic-video redundant sensing module completes the identification of defects, meteorological classification, and traffic flow statistics. The multi-source fusion dynamic risk management module performs risk assessment, level classification, and early warning issuance based on the statistical data.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation, and in particular to a multi-level integrated sensing system and method for monitoring and early warning of highway infrastructure. Background Technology

[0002] In the field of intelligent transportation, highway infrastructure has long faced the dual challenges of complex environment and operational pressure. In particular, some regional highways frequently suffer from road surface wear and bridge and tunnel structural damage due to heavy-load transportation tasks. Coupled with risk factors such as mining subsidence areas, special soil conditions and extreme weather, the probability of infrastructure disasters has increased significantly.

[0003] The existing monitoring system has significant shortcomings: First, monitoring data is isolated, and there is insufficient coordination between edge sensing, edge analysis, and cloud decision-making, resulting in information silos and delayed anomaly detection. Second, it relies excessively on roadside video surveillance and manual inspections, leading to high maintenance costs and low response efficiency. Third, fiber optic resources along the route are only used for communication purposes, and their distributed sensing potential has not been developed, resulting in low resource utilization. Fourth, in long-line highway scenarios, random disasters such as local settlement and hidden cracks are difficult to handle in a timely manner due to monitoring blind spots and response delays, making it difficult to meet the actual needs of full life-cycle safety monitoring and intelligent early warning for highways. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-level integrated sensing system and method for monitoring and early warning of highway infrastructure, which aims to meet the actual needs of safety monitoring and intelligent early warning throughout the entire life cycle of highways.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a multi-level integrated sensing and sensing highway infrastructure monitoring and early warning system. The system includes: an end-side optimized deployment module, a fiber-optic-video dual-channel transmission module, a fiber-optic physical field inversion module, a fiber-optic-video redundant sensing module, and a multi-source fusion dynamic risk management module. The output of the end-side optimized deployment module is connected to the fiber optic signal input and video signal input of the fiber-optic-video dual-channel transmission module, respectively, for sending the collected fiber optic sensing signals and video monitoring data to the fiber-optic-video dual-channel transmission module. The fiber optic signal output of the fiber-optic-video dual-channel transmission module is connected to the input of the fiber-optic physical field inversion module, and the video signal output is connected to the video data input of the fiber-optic-video redundant sensing module, respectively transmitting processed optical signals. The system integrates fiber optic sensing signals and video surveillance data. The output of the fiber optic physical field inversion module is connected to the physical parameter input of the fiber optic-video redundancy sensing module. This conversion transforms the fiber optic sensing signals into physical parameter information, which is then sent to the fiber optic-video redundancy sensing module. The physical parameter information includes stress, strain, vibration, and temperature parameters. The output of the fiber optic-video redundancy sensing module is connected to the input of the multi-source fusion dynamic risk management module. Based on the physical parameter information and video surveillance data, the fiber optic-video redundancy sensing module performs disease identification, meteorological classification, and traffic flow statistics, sending the statistical data to the multi-source fusion dynamic risk management module. The multi-source fusion dynamic risk management module uses the statistical data to perform risk assessment, level classification, and early warning issuance.

[0007] The end-side optimized deployment module includes a fiber optic sensing deployment unit and a video surveillance deployment unit. The fiber optic sensing deployment unit lays optical cables along the existing infrastructure of the highway to form a continuous linear sensing network, and the output end of the fiber optic sensing deployment unit is connected to the fiber optic signal input end. The video surveillance deployment unit adopts a layered, fixed-point, and cross-coverage deployment strategy to set up high-resolution cameras to construct a discrete point-like sensing network, taking into account the different structural characteristics and risk distribution of slopes, bridges, tunnels, and main road sections. The output end of the video surveillance deployment unit is connected to the video signal input end.

[0008] The fiber optic-video dual-channel transmission module includes a fiber optic signal transmission unit and a video signal transmission unit. The input end of the fiber optic signal transmission unit is connected to the fiber optic sensing deployment unit. It receives fiber optic sensing signals and performs modulation, combining, amplification, and relay amplification processing. The output end of the fiber optic signal transmission unit is connected to the input end of the fiber optic physical field inversion module to send the processed fiber optic sensing signals. The input end of the video signal transmission unit is connected to the video surveillance deployment unit, and the output end of the video signal transmission unit is connected to the video data input end of the fiber optic-video redundancy sensing module to send the processed video surveillance data.

[0009] The fiber optic physical field inversion module includes a signal splitting unit and a physical parameter conversion unit. The input of the signal splitting unit is connected to the output of the fiber optic signal transmission unit, receiving the fiber optic sensing signal and splitting it into a main optical path signal and a sensing optical path signal. The output of the sensing optical path signal of the signal splitting unit is connected to the input of the physical parameter conversion unit. The output of the physical parameter conversion unit is connected to the physical parameter input of the fiber optic-video redundancy sensing module. The physical parameter conversion unit can be used to perform photoelectric conversion, noise filtering, amplification, and parameter calculation on the sensing optical path signal.

[0010] The fiber optic-video redundancy sensing module includes a deep defect identification unit, an apparent defect identification unit, a defect matching unit, and a traffic environment monitoring unit. The input of the deep defect identification unit is connected to the output of the fiber optic physical field inversion module, and its output is connected to the first input of the defect matching unit. This unit receives physical parameter information, detects deep defects using a neural network model, and sends the deep defect information to the defect matching unit. The input of the apparent defect identification unit is connected to the output of the video signal transmission unit, and its output is connected to the second input of the defect matching unit. This unit receives video surveillance data and detects defects using a target detection model. The system detects apparent defects and sends the apparent defect information to the defect matching unit. The output of the defect matching unit is connected to the defect information input of the multi-source fusion dynamic risk management module. The defect matching unit is configured to complete defect matching and update the anomaly probability, and send the matched defect data to the multi-source fusion dynamic risk management module. The input of the traffic environment monitoring unit is connected to the output of the video signal transmission unit, and the output of the traffic environment monitoring unit is connected to the environmental traffic information input of the multi-source fusion dynamic risk management module. It is used to receive video monitoring data, identify weather conditions, and count traffic flow, and send the weather condition and traffic flow data to the multi-source fusion dynamic risk management module.

[0011] The multi-source fusion dynamic risk management module includes a risk scoring unit, a risk grading unit, and an early warning release unit. The input of the risk scoring unit is connected to the output of the fiber-optic-video redundant sensing module, and the output of the risk scoring unit is connected to the input of the risk grading unit. It is used to receive disease data, weather conditions, and traffic flow data, calculate the risk score by combining it with location-sensitive information, and send the risk score to the risk grading unit. The output of the risk grading unit is connected to the input of the early warning release unit. It classifies the risk level according to the built-in preset threshold and sends the risk level information to the early warning release unit. The early warning release unit generates differentiated handling strategies based on the regional type and sends early warning information to the management end, the on-site end, and the driver / passenger end.

[0012] Secondly, this invention provides a multi-level integrated sensing method for monitoring and early warning of highway infrastructure, comprising: S1: deploying fiber optic sensing devices and video surveillance devices through an end-side optimized deployment module to collect fiber optic sensing signals and video surveillance data from key areas of highway infrastructure, and sending them to a fiber optic-video dual-channel transmission module; S2: the fiber optic-video dual-channel transmission module performs low-loss transmission of the received fiber optic sensing signals and video surveillance data, sending the processed fiber optic sensing signals to a fiber optic physical field inversion module, and sending the processed video surveillance data to a fiber optic-video redundancy sensing module; S3: the fiber optic physical field inversion module converts the received fiber optic sensing signals into physical parameter information and sends it to the fiber optic-video redundancy sensing module, which combines the physical parameter information and video surveillance data to complete defect identification, meteorological classification, and traffic flow statistics, and sends the results to a multi-source fusion dynamic risk management module; S4: the multi-source fusion dynamic risk management module calculates risk scores, classifies risk levels, and issues differentiated early warning strategies based on the received defect identification results, meteorological conditions, and traffic flow data.

[0013] In step S1, the end-side optimization deployment module lays optical cables along the existing infrastructure of the highway to form a continuous linear sensing network; and adopts a layered, fixed-point and cross-coverage deployment strategy to set up high-resolution cameras to construct a discrete point-like sensing network, taking into account the different structural characteristics and risk distribution of slopes, bridges, tunnels and main road sections.

[0014] In step S3, the signal splitting unit of the fiber optic physical field inversion module splits the received fiber optic sensing signal into a sensing optical path signal. The physical parameter conversion unit performs photoelectric conversion, noise filtering, amplification, and parameter calculation on the sensing optical path signal to generate physical parameter information and send it to the fiber optic-video redundancy sensing module. The fiber optic-video redundancy sensing module detects deep-seated defects based on the physical parameter information and detects apparent defects based on video surveillance data. It converts the coordinates of the two types of defects to a unified coordinate system and completes the matching, updates the anomaly probability, and simultaneously identifies the weather conditions and counts traffic flow based on the video surveillance data. It then sends the defect matching results, weather conditions, and traffic flow data to the multi-source fusion dynamic risk management module.

[0015] In step S4, the multi-source fusion dynamic risk management module calculates the risk score for each area of ​​the highway by combining the severity of the damage, weather conditions, traffic flow, location-sensitive information, and the probability of damage anomalies; it divides the risk score into different risk levels according to preset thresholds; and it generates differentiated handling strategies by combining risk levels and area types, and sends early warning information to the management end, the on-site end, and the driver and passenger end.

[0016] Compared to existing technologies, the beneficial effects of this application are as follows:

[0017] 1. This invention employs a three-tiered modular design. After the edge-side optimized deployment module collects data, it efficiently transmits the data to the fiber optic physical field inversion module and the fiber optic-video redundant sensing module via a fiber optic dual-channel transmission module. Finally, the multi-source fusion dynamic risk management module completes the decision-making process. This breaks the problem of isolated data between the end, edge, and cloud in traditional monitoring, achieving end-to-end collaboration from data collection, transmission, analysis to early warning, significantly improving the timeliness of anomaly detection and decision-making efficiency. Furthermore, the edge-side optimized deployment module utilizes existing communication pipelines, trenches, and other infrastructure along highways to lay optical cables, avoiding redundant construction. It uses optical fibers as both distributed sensors and communication media, solving the resource waste problem of traditional optical fibers being used only for communication, reducing the dense deployment of high-cost video equipment, and significantly lowering construction and long-term maintenance costs.

[0018] 2. The fiber optic-video redundancy sensing module employs a dual-dimensional detection approach, combining deep defect identification (based on physical parameters) and apparent defect identification (based on video data) with a defect matching unit to achieve accurate correlation and anomaly probability updates between the two types of defects, effectively avoiding false detections and missed detections inherent in single monitoring methods. The multi-source fusion dynamic risk management module integrates defect data, weather conditions, traffic flow, and location-sensitive information, achieving accurate risk assessment through risk scoring and grading. Simultaneously, it generates differentiated handling strategies based on regional types (road sections, bridges, tunnels, slopes), issuing appropriate early warning information to management, on-site, and driver / passenger terminals. This ensures both highway structural safety and traffic safety while flexibly adjusting handling timing based on traffic flow, balancing safety and traffic efficiency, and meeting the needs of full life-cycle highway safety monitoring and intelligent early warning. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram of the structure of a multi-level integrated sensing highway infrastructure monitoring and early warning system provided in this application embodiment;

[0021] Figure 2 A schematic diagram of another multi-level integrated sensing highway infrastructure monitoring and early warning system provided in this application embodiment;

[0022] Figure 3 A schematic diagram of a signal amplifier circuit provided in an embodiment of this application;

[0023] Figure 4 A road defect matching result diagram provided in an embodiment of this application;

[0024] Figure 5 A tunnel defect matching result diagram provided in an embodiment of this application;

[0025] Figure 6 A comparison chart of experimental results provided for an embodiment of this application;

[0026] Figure 7 Another experimental result comparison chart provided for an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0028] In the description of the invention, it should be understood that the terms "upper," "lower," "left," "right," "front," "rear," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or relative positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Unless otherwise specified, the above-mentioned orientational descriptions can be flexibly set in practical applications, provided that the relative positional relationships shown in the accompanying drawings are satisfied.

[0029] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0030] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "communication" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. They can refer to a direct connection or an indirect connection through an intermediate medium, or a communication between the internal components of two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0031] In embodiments of the invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, article, or apparatus that includes that element.

[0032] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0033] This application provides a multi-level integrated sensing highway infrastructure monitoring and early warning system, exemplified by, for example... Figure 1 and Figure 2 As shown, the system includes: an end-side optimized deployment module, a fiber-to-video dual-channel transmission module, a fiber-to-optical physical field inversion module, a fiber-to-optical video redundancy sensing module, and a multi-source fusion dynamic risk management module. The output of the end-side optimized deployment module is connected to the fiber optic signal input and video signal input of the fiber-to-optical video dual-channel transmission module, respectively, to send the acquired fiber optic sensing signals and video monitoring data to the fiber-to-optical video dual-channel transmission module. The fiber optic signal output of the fiber-to-optical video dual-channel transmission module is connected to the input of the fiber-to-optical physical field inversion module, and the video signal output is connected to the video data input of the fiber-to-optical video redundancy sensing module, respectively transmitting the processed fiber optic sensing signals and video monitoring data. The output of the physical field inversion module is connected to the physical parameter input of the fiber-optic video redundancy sensing module. This module converts fiber optic sensing signals into physical parameter information and sends this information to the fiber-optic video redundancy sensing module. The physical parameter information includes stress, strain, vibration, and temperature parameters. The output of the fiber-optic video redundancy sensing module is connected to the input of the multi-source fusion dynamic risk management module. Based on the physical parameter information and video surveillance data, the fiber-optic video redundancy sensing module performs disease identification, meteorological classification, and traffic flow statistics, and sends the statistical data to the multi-source fusion dynamic risk management module. The multi-source fusion dynamic risk management module uses the statistical data to perform risk assessment, level classification, and early warning issuance.

[0034] The end-side optimized deployment module includes a fiber optic sensing deployment unit and a video surveillance deployment unit. The fiber optic sensing deployment unit lays optical cables along the existing infrastructure of the highway to form a continuous linear sensing network. The output end of the fiber optic sensing deployment unit is connected to the fiber optic signal input end. The video surveillance deployment unit adopts a layered, fixed-point, and cross-coverage deployment strategy to set up high-resolution cameras to construct a discrete point-like sensing network, taking into account the different structural characteristics and risk distribution of slopes, bridges, tunnels, and main road sections. The output end of the video surveillance deployment unit is connected to the video signal input end.

[0035] As one possible approach, the optical cable is a 12-core GYSTA loose-tube stranded fully filled optical cable conforming to the ITU-T G.652 standard, which has excellent tensile strength, rodent resistance and corrosion resistance. The outer sheath is protected by a PE sheath and filled with moisture-proof grease.

[0036] In slope sections, the optical cable is buried along the outer edge of the drainage ditch at the toe of the slope (on the side closest to the mountain), reusing existing cabling channels in the area. The burial depth is no less than 80 cm, and a 10 cm thick layer of medium-coarse sand is pre-laid at the bottom of the trench, with a 0.5% allowance. At bridge locations, the existing DN50 hot-dip galvanized steel pipes inside the crash barriers are used for cable laying. The pipe wall thickness is no less than 3.0 mm, the bending radius is greater than 600 mm, both ends are waterproofed and sealed, and flexible fireproof sealant is filled between the pipe and the optical cable. In tunnels, the optical cable utilizes existing cable trays or fixed points in the tunnel side ditches. In the main line section, the optical cable is directly buried below the central median strip, laid in existing communication pipes or trenches in the area, with a burial depth of no less than 80 cm. A 5 cm layer of fine sand is laid at the bottom of the trench, and warning tape is continuously placed over the optical cable. When crossing structures, pre-buried Φ50 HDPE protective pipes are used for protection, with both ends of the protective pipe extending at least 1.0 meter beyond the structure.

[0037] High-resolution cameras are deployed in layers along the slope: from the top to the middle and then to the bottom, facing each other and at least 2 meters from the bottom. On bridges, two video surveillance devices are installed at the bottom of the mid-span beams, with lenses pointing vertically upwards to cover the entire width of the beam bottom. In tunnels, high-definition cameras are added at approximately 500-meter intervals along the tunnel length, either on the side walls or above maintenance pipelines. Deployment density is increased in long tunnels / curves, near important equipment boxes / power distribution rooms, key fire-fighting points, and disaster-prone areas. The main line section adopts a "central + roadside" cross-coverage pattern, with high-resolution cameras deployed at a density of approximately 1000 meters per unit. This deployment pattern, combined with the aforementioned linear continuous coverage fiber optic sensor network, forms a complementary point-line sensing structure with no blind spots across the entire area.

[0038] The fiber optic-video dual-channel transmission module includes a fiber optic signal transmission unit and a video signal transmission unit. The input end of the fiber optic signal transmission unit is connected to the fiber optic sensing deployment unit. It receives fiber optic sensing signals and performs modulation, combining, amplification, and relay amplification processing. The output end of the fiber optic signal transmission unit is connected to the input end of the fiber optic physical field inversion module to send the processed fiber optic sensing signals. The input end of the video signal transmission unit is connected to the video surveillance deployment unit, and the output end of the video signal transmission unit is connected to the video data input end of the fiber optic-video redundancy sensing module to send the processed video surveillance data.

[0039] The fiber optic-video dual-channel transmission module achieves low-loss and high-fidelity recovery of long-distance fiber optic signals during transmission through a dynamic low-loss transmission method, ensuring stable transmission and accurate decoding of detection and communication data. For key monitoring points, an independent video transmission link is established, dedicated to transmitting high-definition video streams. This forms a dual-channel transmission mode at the system level, with primary and backup complementarity and each having its own focus, comprehensively ensuring the reliability, real-time performance, and stability of various monitoring data transmissions on highways.

[0040] It should be understood that the input end of the fiber optic signal transmission unit is also the fiber optic signal input end, and the output end of the fiber optic signal transmission unit is also the fiber optic signal output end; the input end of the video signal transmission unit is also the video signal input end, and the output end of the video signal transmission unit is also the video signal output end.

[0041] For example, the transmitter of the fiber optic signal transmission unit is deployed in a standard communication room at a highway toll station or roadside monitoring center. A narrow-linewidth tunable laser injects probe light into the buried single-mode fiber along the route. The Brillouin scattering signal generated by the fiber is collected, isolated by a circulator in the signal splitter unit, and then fed into a coherent receiving module. A photoelectric conversion is performed by a photodetector to form raw detection data. After analog-to-digital conversion, the raw monitoring data is input into a dual-branch hybrid neural network to compensate for digital pre-distortion. This network extracts the spatiotemporal features and phase-amplitude implicit characteristics of the signal through real and complex branches, respectively. The real branch, based on a one-dimensional convolutional neural network and an attention mechanism, receives a 2S-dimensional real-valued vector after I / Q separation, extracts temporal and spatial features through convolutional layers, and introduces channel attention and spatial attention mechanisms to enhance the perception of key damage features. The calculation formula is as follows:

[0042]

[0043] Where Attn represents the attention module, This represents the convolution block operation. This is the weight matrix. For constant bias terms, is the feature vector output by the real branch. r is the real branch signal. The complex branch uses a fully connected complex neural network to model the amplitude and phase changes of the signal, directly processing s complex symbols. It uses a complex activation function based on the Kerr effect to better model the nonlinear phase change of the light field. The calculation formula is as follows:

[0044]

[0045] in, Let be the complex input vector of the k-th layer neuron. This is the weight matrix. For constant bias terms, The resurrection activation function used for the k-th layer.

[0046] The outputs of the two branches are concatenated at the feature layer and then input into the fully connected layer for joint recovery. The calculation formula is as follows:

[0047]

[0048] In this context, ReBranch and ImBranch represent the real and complex branches, respectively, and W and b are the final linear layer parameters. This is the recovered data signal.

[0049] The detection data processed by the dual-branch hybrid neural network is preprocessed and compressed, and then remodulated into an optical signal using dual-polarization differential phase-shift keying (DP-DPSK) modulation. This allows information to be carried using the phase difference between adjacent symbols, and the symbol mapping relationship can be expressed as:

[0050]

[0051] in, Represents the complex value of the k-th symbol. The differential coded bits to be modulated are currently being used. The complex value of the (k-1)th symbol represents the symbol state at the previous moment. After modulation, the same signal is transmitted simultaneously and independently through two orthogonal polarization states, x-polarization and y-polarization. The resulting DP-DPSK optical signal can be expressed as:

[0052]

[0053] in, To output the complex amplitude of the light field, and These represent the time-domain envelopes of the electric field in the x and y polarization directions, respectively. The optical carrier angular frequency, and These are the initial phases in the two polarization directions, respectively.

[0054] The wavelength division multiplexer (WDM) combines modulated optical signals of different wavelengths into a composite optical signal, which is then amplified by an erbium-doped fiber amplifier (EDFA) and injected into the fiber optic trunk for transmission.

[0055] During transmission, relay amplification stations are set up along highway service areas or tunnel management station equipment rooms. Each station is equipped with an EDFA (Electronic Digital Amplifier) ​​to amplify the optical signal, compensating for power attenuation caused by long-distance transmission. Distributed video acquisition terminals are deployed in key monitoring areas of the highway, and video stream data is directly transmitted through independent video transmission links to ensure the real-time performance and reliable transmission of critical monitoring videos.

[0056] The fiber optic physical field inversion module includes a signal splitting unit and a physical parameter conversion unit. The input of the signal splitting unit is connected to the output of the fiber optic signal transmission unit, receiving the fiber optic sensing signal and splitting it into a main optical path signal and a sensing optical path signal. The output of the sensing optical path signal of the signal splitting unit is connected to the input of the physical parameter conversion unit. The output of the physical parameter conversion unit is connected to the physical parameter input of the fiber optic-video redundancy sensing module. The physical parameter conversion unit can be used to perform photoelectric conversion, noise filtering, amplification, and parameter calculation on the sensing optical path signal.

[0057] The signal splitting unit receives the fiber optic sensing signal output from the fiber optic signal transmission unit and inputs it into a three-port non-reciprocal fiber optic circulator. The three-port non-reciprocal fiber optic circulator splits the incident fiber optic sensing signal into two parts: a main optical path signal and a sensing optical path signal. The main optical path signal has a transmittance significantly higher than 95%, ensuring low-loss input into the backbone sensing fiber for transmission, used to transmit the fiber optic sensing signal to subsequent sensing fibers and ensure the continuity of the sensing link. The sensing optical path signal has a coupling efficiency better than 90% and is directionally transmitted to the physical parameter conversion unit to provide the raw signal basis for obtaining strain, stress, vibration, and temperature parameter information.

[0058] For example, the core component of the physical parameter conversion unit is a distributed fiber optic sensing unit, which includes a photoelectric conversion circuit, a signal amplifier circuit, and a physical parameter conversion circuit. The photoelectric conversion circuit receives the sensing optical path signal output from the signal splitter unit and converts it into an initial electrical signal. The signal amplifier circuit receives the initial electrical signal output from the photoelectric conversion circuit, performs noise filtering and amplification on the signal, eliminates signal interference, and enhances signal strength, providing a high-quality electrical signal for subsequent parameter calculations. The physical parameter conversion circuit performs parameter calculations on the amplified electrical signal and finally outputs strain, stress, vibration, and temperature parameter information, which is then output to the physical parameter input terminal of the fiber-optic-video redundant sensing module.

[0059] As one possible implementation, such as Figure 3 As shown, the signal amplifier circuit includes an input voltage divider network, an operational amplifier U1pA, an optocoupler, a feedback network, and a noise suppression unit. The input voltage divider network is composed of a first resistor R1 and a fifth resistor R5. The input terminal (CHECK 11) of the signal amplifier circuit is grounded through R1. R1 and R5 form a voltage divider branch to achieve voltage adjustment. The inverting input terminal of U1pA is connected to a capacitor C1 and a resistor R2 in sequence, and the non-inverting input terminal of U1pA is connected to a positive voltage VCC_5P. The optocoupler consists of a light-emitting diode (LED) P1 and a photosensitive diode (Q1). The cathode of P1 is connected to the output terminal of U1pA via resistor R3, and the anode of P1 is grounded. The cathode of Q1 is connected to the output terminal of the signal amplifier circuit (ADCHP 11), and the anode of Q1 is grounded. The positive voltage VCC_5P of the feedback network is connected to the voltage divider node of R1 and R5 via a reverse diode D2. The anode of the reverse diode D1 is grounded to GND, and the cathode of the reverse diode D1 is connected to the anode of the reverse diode D2. One end of capacitor C1 is connected to the voltage divider node of R1 and R5, and the other end is connected to the inverting input terminal of U1pA via R2. The output terminal of the signal amplifier circuit is connected to the voltage divider node of R1 and R5 via capacitor C3, and simultaneously connected to the series node of C1 and R2 via resistor R4. The noise suppression unit filters out power supply noise by connecting capacitor C2 in parallel between VCC_5P and ground.

[0060] The fiber optic-video redundancy sensing module includes a deep defect identification unit, an apparent defect identification unit, a defect matching unit, and a traffic environment monitoring unit. The input of the deep defect identification unit is connected to the output of the fiber optic physical field inversion module, and its output is connected to the first input of the defect matching unit. This unit receives physical parameter information, detects deep defects using a neural network model, and sends the deep defect information to the defect matching unit. The input of the apparent defect identification unit is connected to the output of the video signal transmission unit, and its output is connected to the second input of the defect matching unit. This unit receives video surveillance data and detects defects using a target detection model. The system detects apparent defects and sends the apparent defect information to the defect matching unit. The output of the defect matching unit is connected to the defect information input of the multi-source fusion dynamic risk management module. The defect matching unit is configured to complete defect matching and update the anomaly probability, and send the matched defect data to the multi-source fusion dynamic risk management module. The input of the traffic environment monitoring unit is connected to the output of the video signal transmission unit, and the output of the traffic environment monitoring unit is connected to the environmental traffic information input of the multi-source fusion dynamic risk management module. It is used to receive video monitoring data, identify weather conditions, and count traffic flow, and send the weather condition and traffic flow data to the multi-source fusion dynamic risk management module.

[0061] For example, the deep defect identification unit uses the KAST-Net neural network model to analyze the physical parameter information output by the fiber optic physical field inversion module to identify deep defects in highway infrastructure.

[0062] As one possible approach, the KAST-Net neural network model is used to analyze the physical parameter information output by the fiber optic physical field inversion module to identify deep-seated problems in highway infrastructure. The specific operations include: fusing the physical parameter information of each location along the fiber optic line within a specified time interval (time t to tk), and combining it with the structural location information corresponding to that location, such as the specific section location of slopes, bridges, and tunnels, to construct a multi-channel spatiotemporal map.

[0063] To achieve spatiotemporal modeling of infrastructure conditions, this application embodiment simultaneously acquires strain, temperature, and vibration signals (spatial resolution approximately 1m, sampling interval approximately 10s). In practical applications, considering a total line length of approximately 5km, the time window is set to 1 minute. Under this configuration, each time window contains approximately 6 frames of strain data, 30 frames of temperature data, and corresponding vibration signal data. After time alignment and resampling, these are uniformly formed into 60 frames of time-series samples. The spatial dimension is discretized at a 1m resolution, resulting in a total of 5000 spatial points. Based on this, [the following is a continuation of the previous sentence, but the translation is incomplete]. The strain-location data, stress data converted from the material's elastic modulus, temperature-location data, and vibration signal data are used as input channels, and structural location information is introduced (roads, bridges, tunnels, and slopes are represented by one-hot encoding). Finally, a multi-channel spatiotemporal map with a size of 8×60×5000 is constructed.

[0064] Multi-layer depthwise separable convolutions are employed to extract features from multi-channel spatiotemporal maps. Lightweight convolution operations preserve key features and reduce computational cost. A temporal dynamic attention module is introduced in the final layer of the depthwise separable convolutions. This module performs global average pooling, global max pooling, and variance pooling operations on the features at each time step along both the spatial and channel dimensions. The pooling results are concatenated and input into a fully connected layer, then normalized using the Softmax function to generate temporal attention weights. These weights are used to weight and scale the features at each time step. Finally, a Feature Pyramid Network (FPN) fuses the extracted high- and low-level features. High-level features correspond to global trend information of the disease, while low-level features correspond to local details. This fusion improves the completeness and accuracy of deep disease features, providing high-quality feature data for subsequent disease identification.

[0065] The calculation process is as follows:

[0066]

[0067]

[0068]

[0069] in, This represents the feature map at time step t. , , These represent global average pooling, max pooling, and variance pooling operations, respectively. This represents the vector obtained by concatenating the pooled features. This represents the fully connected layer used to generate attention weights. This indicates the Softmax normalization operation. This represents the attention weight corresponding to the t-th time step. This represents the feature representation after attention weight scaling.

[0070] For example, the feature extraction module uses three depthwise separable convolutional layers for initial feature extraction, with each layer having a 5×5 kernel. The number of output channels increases proportionally by 2: the first layer expands from 8 input channels to 16 channels and downsamples the spatial dimension with a stride of 2, resulting in an output size of 16×60×2500; the second layer expands from 16 channels to 32 channels and further downsamples the spatial dimension to 1250; the third layer expands from 32 channels to 64 channels and downsamples the spatial dimension to 625. Each convolutional layer is followed by batch normalization and a ReLU activation function to enhance feature representation and suppress overfitting.

[0071] A time-dimensional dynamic attention module is introduced after the third convolutional layer: global average pooling, max pooling and variance pooling are performed on the 60 time-step features respectively to obtain three 60-dimensional vectors, which are then concatenated and input into two fully connected layers to generate 60 time weights. The weight coefficients are normalized by softmax and used to perform weighted scaling on the features of each time step.

[0072] In addition, a feature pyramid network is introduced into the module: First, the 64-channel feature of the high layer (60×625) is upsampled to the same size as the 32-channel feature of the middle layer (60×1250) and stitched together to obtain a 96-channel feature map; then it is upsampled to be aligned with the 16-channel feature of the low layer (60×2500) and stitched together to obtain a 112-channel feature map, which is the final output (112×60×2500).

[0073] The input feature map, derived from the multi-channel spatiotemporal features of physical parameter information and processed as described above, is divided into several independent feature blocks using a KAN (Kernel Additive Networks)-Transformer encoder layer. Each feature block is then reshaped into a fixed-dimensional vector using a fully connected layer, and positional encoding is added to generate an output vector containing positional information. The specific calculation process is as follows:

[0074]

[0075] in, This represents the i-th feature block obtained from the partitioning. This indicates the feature transformation performed by the fully connected layer. This represents the position code at the corresponding location. This represents the i-th output vector after adding position information. Next, all... Combined into a sequence The input is fed into a multi-layered stacked Transformer encoder. Each encoder layer consists of a multi-head self-attention module (MHSA) and a feedforward network based on a KAN structure. The specific calculation process is as follows:

[0076]

[0077] in, This indicates a bullish self-attention strategy. This represents a feedforward transform based on the KAN structure. This represents the feature sequence output by the l-th layer encoder.

[0078] For example, the fused feature map output by the feature extraction module is 112×60×2500. This map is divided into non-overlapping blocks along the temporal and spatial dimensions, with each block measuring 112×4×10. This results in 15 segments (60 / 4) in the temporal direction and 250 segments (2500 / 10) in the spatial direction, yielding a total of 3750 blocks. Each block is flattened and reshaped into a 256-dimensional vector by a fully connected layer, forming a vector sequence of length 3750.

[0079] Sine / cosine positional encoding (with an encoding dimension of 256, and the positional information includes both temporal and spatial location) is added to this vector sequence and fed into a network consisting of four stacked Transformer encoder layers. Each layer contains an eight-head self-attention mechanism, and its feedforward network is replaced by KAN to improve the spatiotemporal feature modeling capability and nonlinear fitting effect.

[0080] The reconstructed output layer uses convolution operations to upsample the upper-layer features output from the KAN-Transformer encoder layer layer by layer, restoring them to the level of the input features of the deep disease identification unit. Figure 1 The dimensions (C, T, L) are determined, and the spatial features at the corresponding time step t are extracted. This method accurately correlates the abnormal physical parameters of each monitoring location at a specific time point. It maps these parameters to a tensor of size (K,L) through a fully connected layer, where K is the number of disease types to be detected. Finally, a sigmoid activation function is used to output the anomaly probability. The specific calculation process is as follows:

[0081]

[0082] in and These represent the weight matrix and bias term of the fully connected layer, respectively. Let represent the Sigmoid activation function, and P represent the output anomaly probability matrix.

[0083] For example, the Transformer encoder outputs 3750 256-dimensional vectors. These vectors are first mapped back to 4480 dimensions through a fully connected layer and restored to small blocks of (112×4×10) shape. These blocks are then de-patched and concatenated to obtain a complete 112×60×2500 feature map. Subsequently, they are progressively restored and compressed through three convolutional layers. Based on this, the spatial dimension is restored to its original resolution through bilinear upsampling, resulting in an 8×60×5000 multi-channel spatiotemporal feature map. The spatial features (8×1×5000) at the current time step t are further extracted, mapped to a (K×L) tensor through a fully connected layer, and the anomaly probability is output through a Sigmoid activation function. Anomaly categories include: road settlement, road cracks, tunnel cracking, tunnel deformation, bridge cracks, abnormal bridge cable tension, abnormal bridge deflection, slope slippage, slope rockfall, and fiber optic breakage.

[0084] For example, this application uses experimental monitoring data from the Shanxi Provincial Expressway Network as the dataset to compare the KAST-Net method with traditional and mainstream spatiotemporal detection methods, conducting experiments on the monitoring effect of deep-seated defects. The experimental monitoring data from the Shanxi Provincial Expressway Network covers four types of infrastructure: roads, bridges, tunnels, and slopes, including 12,000 samples of deep-seated defects. The hardware environment was an Intel Core i5-10400F CPU + NVIDIA GTX 1650 GPU. The experimental results are as follows... Figure 6 As shown, according to Figure 6 It can be seen that the method provided in this application embodiment has the highest accuracy, the shortest inference time, the highest F1 score, and the highest recall rate, which is better than all the comparison methods and meets the real-time monitoring requirements.

[0085] To verify the function of each core component of KAST-Net, an ablation experiment was designed, for example, such as... Figure 7 As shown. Depthily Separable Convolution (DWC): Removing it reduced accuracy by 3.1%, but it retains key features while reducing computational cost, making it a core guarantee of model efficiency. Temporal Dynamic Attention (TDA): Removing it reduced accuracy by 3.5%, but it can focus on key time steps (such as sudden changes in disease progression), improving the utilization rate of temporal features. FPN Structure: Removing it reduced accuracy by 4%, confirming the importance of multi-scale feature fusion in capturing disease features at different spatial scales. KAN Feedforward Network: Replacing it with an MLP reduced accuracy by 4.8%, demonstrating its nonlinear fitting advantages and better adaptability to complex sensor data distributions. Transformer Encoder: Removing it reduced accuracy by 16.2%, validating its ability to model long-distance spatiotemporal dependencies, making it a core component for improving the accuracy of deep disease detection.

[0086] The apparent defects identification unit uses a target detection model, such as the YOLOv12 model, to process video surveillance data output from the fiber optic-video dual-channel transmission module to identify apparent defects in infrastructure. The YOLOv12 target detection model is trained and fine-tuned on training data. During training, model parameters and loss function weights are continuously adjusted, and various data augmentation strategies (such as random cropping, rotation, scaling, and brightness and contrast adjustments) are introduced to enhance the model's adaptability to different road segment scenarios, thereby improving its anomaly detection performance. The goal of fine-tuning is to ensure that the model can accurately identify various types of apparent defects while maintaining robustness to different lighting, weather, and traffic conditions. After training, the optimized YOLOv12 model is deployed to the edge computing unit for real-time video stream processing.

[0087] The disease matching unit constructs an association matching probability model that considers location errors, and performs spatial coordinate matching between detected deep-seated diseases and apparent diseases to ensure the spatiotemporal correlation of the two types of disease information. For example, Figure 4 and Figure 5 As shown, Figure 4 and Figure 5 The matching results of road defects and tunnel defects are shown respectively. Figure 4 and Figure 5 It can be seen that road defects include cracks, deep settlement, and potholes, while tunnel defects include cracks and deep cracking.

[0088] The traffic environment monitoring unit uses classification models, such as the ConvNeXt model, to analyze video surveillance data, classify weather conditions, and employs the TransTrack model to track and count vehicles in the video, achieving traffic flow statistics. For specific road sections, environmental information is processed. First, the ConvNeXt model is used to classify the collected video or image data according to weather conditions. Data preprocessing includes image cropping, scaling, normalization, and illumination equalization. A dataset covering categories such as sunny, cloudy, rainy, snowy, and foggy days is constructed, and video frames are labeled. Based on this dataset, the ConvNeXt model is trained and fine-tuned to adapt to the shooting angle, lighting conditions, and weather environment of specific road sections. After training, it is deployed to the edge computing unit to achieve real-time weather condition determination. Simultaneously, traffic flow statistics are performed on the same video stream. First, the video undergoes denoising, frame rate normalization, and lane segmentation. Then, the TransTrack model is used to track and count vehicles in consecutive frames, achieving real-time traffic flow monitoring and providing auxiliary information for traffic management and anomaly detection.

[0089] In some embodiments, the defect matching unit employs an association matching probability model that considers positional errors. For apparent defects detected by the apparent defect identification unit based on video surveillance data, the association matching probability model considering positional errors maps the pixel coordinates to the camera coordinate system after distortion correction and camera intrinsic parameters, and then combines this with camera extrinsic parameter projection to convert them into three-dimensional spatial coordinates in the structural reference coordinate system. Simultaneously, for deep defects detected by the deep defect identification unit based on physical parameter information, the coordinates of the measurement points along the fiber optic cable length are combined with the actual fiber optic cable layout route in the construction drawings, and mapped to three-dimensional spatial coordinates in the structural reference coordinate system using a piecewise linear interpolation method.

[0090] For example, the process of obtaining coordinates through piecewise linear interpolation includes: obtaining the known coordinates of the start and end points of each fiber optic segment through construction drawings, determining the fiber optic segment i where the defect is located, and recording the relative position of the defect along that fiber optic segment as... The calculation formula is as follows:

[0091]

[0092] in, This indicates the location of the defect in the fiber optic length coordinate system. and These represent the starting and ending points of fiber segment i in the fiber length coordinate system. The three-dimensional spatial coordinates of the defect are obtained by linear interpolation of the three-dimensional spatial coordinates of the starting and ending points of the segment with their relative positions. The calculation formula is as follows:

[0093]

[0094] in, The three-dimensional spatial coordinates of the disease. and These are the three-dimensional spatial coordinates of the starting and ending points of fiber segment i, respectively.

[0095] Then, each pair of deep-seated diseases was calculated. With apparent diseases The spatial Euclidean distance between them is calculated, and the matching probability is obtained based on the distance distribution according to the corresponding calculation logic.

[0096] The calculation formula is as follows:

[0097] in, Indicates disease and The matching probability, Indicates the disease and Calculate the Euclidean distance using the spatial coordinates. Represents an exponential function. The variance parameter represents the distance distribution.

[0098] Finally, a probability threshold is set to remove low-probability matching results. For the remaining candidate matching pairs, the Hungarian algorithm is used to achieve the highest probability one-to-one matching. For diseases that are successfully matched, their original abnormal probability is increased to the updated abnormal probability according to the corresponding update logic. The updated abnormal probability will be used as the core output result of the disease matching unit and further transmitted to the multi-source fusion dynamic risk management module for subsequent risk quantification assessment.

[0099] The update formula is as follows:

[0100]

[0101] in, This represents the raw value of the anomaly probability output by the model. This represents the anomaly probability value updated after matching. This is the probability amplification factor.

[0102] The multi-source fusion dynamic risk management module includes a risk scoring unit, a risk grading unit, and an early warning release unit. The input of the risk scoring unit is connected to the output of the fiber-optic-video redundant sensing module, and the output of the risk scoring unit is connected to the input of the risk grading unit. It is used to receive disease data, weather conditions, and traffic flow data, calculate the risk score by combining it with location-sensitive information, and send the risk score to the risk grading unit. The output of the risk grading unit is connected to the input of the early warning release unit. It classifies the risk level according to the built-in preset threshold and sends the risk level information to the early warning release unit. The early warning release unit generates differentiated handling strategies based on the regional type and sends early warning information to the management end, the on-site end, and the driver / passenger end.

[0103] It should be understood that the input end of the risk scoring unit is connected to the output end of the fiber optic-video redundancy sensing module. The input end of the risk scoring unit is also the environmental traffic information input end and the defect information input end of the multi-source fusion dynamic risk management module. The output end of the fiber optic-video redundancy sensing module is also the output end of the traffic environment monitoring unit and the defect matching unit. The two are connected accordingly.

[0104] For example, the risk scoring unit establishes a dynamic weighting model based on the susceptibility of different areas (road sections, bridges, tunnels, slopes), i.e., the differences in tolerance and response to susceptibility to susceptibility in different areas, to determine the differentiated weights of each area in risk assessment. Furthermore, multiple key factors are introduced in the subsequent risk scoring process: including weather factors (meteorological data) output by the traffic environment monitoring unit, traffic flow factors (vehicle flow data), and the probability of anomalies output by the susceptibility matching unit in the fiber-optic-video redundancy sensing module, forming a multi-factor fusion dynamic risk grading method. The probability of anomalies from the fiber-optic-video redundancy sensing module is used as a core input parameter in the risk scoring formula to accurately calculate the risk scores for each area of ​​the highway. The risk grading unit compares the risk scores calculated by the risk scoring unit with risk thresholds pre-set by expert assessments or obtained through statistical learning from historical monitoring data, ultimately completing the risk level classification. The classification results are further transmitted to the early warning release unit, providing a core basis for generating differentiated response strategies and early warning information.

[0105] For example, the output of the fiber-optic video redundancy sensing module includes the coordinates of the monitoring area. Regional types (Roads, bridges, tunnels, slopes) Abnormal event types Abnormal event physical quantity vector Weather factors Traffic flow factors Through normalization, each physical quantity is transformed into a dimensionless index, combined with a position-sensitive factor. Obtain the equivalent amplitude The severity of abnormal events is obtained by using a piecewise linear saturation function mapping. The calculation steps are as follows:

[0106] Physical quantities of abnormal events Data processing is performed, and the original physical quantity vector is denoted as:

[0107]

[0108] For each physical quantity, perform anchor point normalization:

[0109]

[0110] in: This refers to the kth original physical quantity (e.g., crack length, crack width, deformation angle, etc.) provided in step S4 of the method provided in the embodiments of this application. The upper bound of the normal range for the k-th physical quantity is determined by industry standards or the 95th percentile of historical normal samples, or confirmed by experts. The limit / hazard threshold of the k-th physical quantity is determined by industry standards or historical accident / response samples, or confirmed by experts. This is considered the limiting state; This represents a truncation function that restricts x to [0,1]. This represents the dimensionless value (0–1) of the k-th physical quantity after anchor point normalization.

[0111] Consider location-sensitive information The equivalent amplitude is obtained:

[0112]

[0113]

[0114] in: This indicates location-sensitive information, and the value varies depending on the region, determined by the asset metadata table or expert table. This represents the dominant term after standardization of each physical quantity; k is the ordinal number of the original physical quantity.

[0115] Equivalent amplitude Mapping to severity A is achieved using a piecewise linear monotonic saturation function:

[0116]

[0117] in: This is the normal / concern / limit anchor point corresponding to the equivalent amplitude in this region, set to the default value. , , .

[0118] Specifically, the risk scoring formula is as follows:

[0119]

[0120] in: In This represents the risk score calculated for region j; Represents the coordinates of the region; Indicates the region Types (road sections, bridges, tunnels, slopes); Indicates the region Category of the anomaly ; Indicates the region An anomaly occurred The severity is calculated from the input anomalous physical quantity vector; Represents the dynamic weighting of disease context in the region; Indicates the region Weather conditions; Indicates the region Traffic flow conditions; For reference traffic volume, the designed capacity or historical average traffic volume shall be used; Location-sensitive information is assigned different values ​​depending on the region, determined by the asset metadata table or expert table. Indicates the region An anomaly occurred The probability of; , , , These are the parameters to be optimized.

[0121] Optimize parameters , , , This is obtained by solving a regularized nonlinear least squares optimization problem, specifically including the following steps:

[0122] Collect historical sample data ,in This includes region type, anomaly category, historical weather and traffic flow characteristics. This represents the actual observed risk score.

[0123] Construct the following loss function:

[0124]

[0125] in: This is the regularization coefficient.

[0126] The Levenberg-Marquardt iterative method is used to minimize the loss function and update the parameters. :

[0127]

[0128] in: Represents the Jacobian matrix; It is the damping factor; This indicates that in the nth iteration, the current parameter vector is used. The calculated predicted risk scores for all samples; Represents the observation label vector; It is the identity matrix. Jacobian matrix The transpose of .

[0129] As one possible implementation method, a threshold is set based on expert evaluation. In comparison, the risk grading unit divides the risk level into five levels: 0, 1, 2, 3, and 4, based on the score, corresponding to 0 ≤ R < 4. , ≤R< , ≤R< , ≤R< and R≥ Level 0 is normal (requires recording), Level 1 is low risk (requires marking), Level 2 is medium risk (requires warning and flow restriction), Level 3 is high risk (requires maintenance and partial lockdown), and Level 4 is extremely high risk (requires immediate closure).

[0130] Based on risk level, region type, and traffic flow level, intelligent multi-dimensional differentiated strategies are generated. Strategy principles include:

[0131] (a) Safety priority: personal safety > structural safety > traffic efficiency > cost;

[0132] (b) During high traffic periods, maintenance of low-risk events may be postponed and avoidance notices may be issued; maintenance should be prioritized during low traffic periods.

[0133] (c) For bridges, tunnels and slopes, when the risk level reaches medium or above, the handling should not be delayed due to high traffic volume;

[0134] (d) If the prediction model shows that there are significant risks in the future, even if the current risk level is low, issue key inspection or preventive maintenance instructions in advance;

[0135] (e) The information is distributed to the management end, the inspection / maintenance end and the driver and passenger end. Different ends receive risk warning information of different granularities. Normal information is not distributed and is only used for internal baseline updates.

[0136] Based on traffic flow levels (low, medium, high) and area types (road sections, bridges, tunnels, slopes), differentiated response measures are formulated for the same risk level to achieve multi-dimensional and intelligent early warning strategy issuance, as follows: For road areas, at level 0, regardless of traffic flow, only management records are made, no on-site measures are taken, drivers and passengers drive normally, and monitoring is conducted at the regular frequency; at level 1 (low flow), management marks anomalies and generates inspection tasks, on-site observation is marked, drivers and passengers are reminded to pay attention, and monitoring is increased to once a day; at medium flow, management generates inspection work orders, conducts on-site inspections, and monitors twice a day; at high flow, management marks defects and postpones inspections to the low flow stage, and monitoring is conducted three times a day; at level 2 (low flow), priority is given to nighttime monitoring. For Level 3 low traffic, immediate short-term closure and repair are implemented, along with on-site traffic control and diversion reminders for drivers and passengers, and monitoring every 6 hours. For Level 4 low traffic, the entire area is closed off, traffic police are notified, emergency repairs are carried out, the area is immediately closed off, drivers and passengers are forced to detour, and monitoring is conducted every 3 hours. For Level 5 high traffic, speed limits are immediately issued, speed limit signs and diversion signs are set up, and monitoring is conducted every hour. For Level 6 low traffic, short-term closure and repair are implemented, on-site isolation and diversion are implemented, drivers and passengers are reminded to detour, and monitoring is conducted every hour. For Level 6 medium traffic, closure and repair are implemented at night, and traffic control is deployed, with monitoring conducted every 30 minutes. For Level 6 high traffic, a complete closure is implemented, emergency teams are dispatched, the area is completely closed off, drivers and passengers are forced to detour, and real-time monitoring is conducted.

[0137] For bridge areas, Level 0 is the same as for roads; for Level 1 (low / medium flow), the frequency of management inspections is increased to once a day, on-site traffic restrictions are marked, drivers and passengers are reminded to pay attention, and monitoring is conducted once a day. For high flow, management inspections and real-time monitoring are strengthened, and monitoring is conducted once every 6 hours; for Level 2 (low flow), load / speed limits are imposed and experts are dispatched for review, on-site weight limit signs are set up, and monitoring is conducted once every 3 hours. For medium flow, expert evaluation and nighttime emergency repairs are added, drivers and passengers are advised to detour, and monitoring is conducted once every 2 hours. For high flow, monitoring is intensified to once every 30 minutes, on-site support and reinforcement are carried out, and some drivers and passengers detour; for Levels 3-4, all areas are completely closed regardless of flow. Level 3 prohibits the passage of heavy-load vehicles, and Level 4 activates the emergency plan, drivers and passengers are forced to detour, and real-time monitoring is conducted.

[0138] Level 0: No difference in tunnel area; Level 1: Low / medium flow inspections increased to once a day, on-site observation, and driver / passenger reminders; high flow monitoring every 6 hours; Level 2: Low flow intermittent release, monitoring every 3 hours; medium flow intermittent release + nighttime emergency repairs and lane closures, monitoring every hour; high flow with high predicted risk: early closure, on-site diversion, drivers / passengers detour in advance, monitoring every 30 minutes; Levels 3-4 (high / extremely high risk): complete closure; Level 3: tunnel closure for evacuation and maintenance; Level 4: activation of major emergency plans, mandatory detours for drivers / passengers, and real-time monitoring.

[0139] Level 0 for slope areas: Level 1: Low / medium flow, intensive monitoring and photography, on-site observation, and monitoring once a day; high flow, additional rain warnings and monitoring every 6 hours; Level 2: Low flow, if rain is predicted, advance traffic restrictions / closures, on-site warning lines, and monitoring every 3 hours; medium flow, traffic restrictions and engineering assessments, and monitoring every hour; high flow, immediate closure if rain is expected, and minor repairs if no rain is expected, and monitoring every 30 minutes; Level 3: Closure, evacuation, reinforcement, and emergency repairs; on-site isolation; detours for drivers and passengers; and real-time monitoring; Level 4: Large-scale evacuation and long-term detour planning; complete on-site closure; mandatory detours for drivers and passengers; and real-time monitoring.

[0140] Secondly, this invention provides a multi-level integrated sensing method for monitoring and early warning of highway infrastructure, comprising: S1: deploying fiber optic sensing devices and video surveillance devices through an end-side optimized deployment module to collect fiber optic sensing signals and video surveillance data from key areas of highway infrastructure, and sending them to a fiber optic-video dual-channel transmission module; S2: the fiber optic-video dual-channel transmission module performs low-loss transmission of the received fiber optic sensing signals and video surveillance data, sending the processed fiber optic sensing signals to a fiber optic physical field inversion module, and sending the processed video surveillance data to a fiber optic-video redundancy sensing module; S3: the fiber optic physical field inversion module converts the received fiber optic sensing signals into physical parameter information and sends it to the fiber optic-video redundancy sensing module, which combines the physical parameter information and video surveillance data to complete defect identification, meteorological classification, and traffic flow statistics, and sends the results to a multi-source fusion dynamic risk management module; S4: the multi-source fusion dynamic risk management module calculates risk scores, classifies risk levels, and issues differentiated early warning strategies based on the received defect identification results, meteorological conditions, and traffic flow data.

[0141] In step S1, the end-side optimization deployment module lays optical cables along the existing infrastructure of the highway to form a continuous linear sensing network; and adopts a layered, fixed-point and cross-coverage deployment strategy to set up high-resolution cameras to construct a discrete point-like sensing network, taking into account the different structural characteristics and risk distribution of slopes, bridges, tunnels and main road sections.

[0142] In step S3, the fiber optic physical field inversion module splits the received fiber optic sensing signal into a sensing optical path signal, performs photoelectric conversion, noise filtering, amplification, and parameter calculation on the sensing optical path signal, generates physical parameter information, and sends it to the fiber optic-video redundancy sensing module. The fiber optic-video redundancy sensing module detects deep-seated defects based on the physical parameter information and detects apparent defects based on video surveillance data. It converts the coordinates of the two types of defects to a unified coordinate system and completes the matching, updates the anomaly probability, and simultaneously identifies the weather conditions and counts traffic flow based on the video surveillance data. It then sends the defect matching results, weather conditions, and traffic flow data to the multi-source fusion dynamic risk management module.

[0143] In step S4, the multi-source fusion dynamic risk management module calculates the risk score for each area of ​​the highway by combining the severity of the damage, weather conditions, traffic flow, location-sensitive information, and the probability of damage anomalies; it divides the risk score into different risk levels according to preset thresholds; and it generates differentiated handling strategies by combining risk levels and area types, and sends early warning information to the management end, the on-site end, and the driver and passenger end.

[0144] The method provided in this application embodiment is based on the above-mentioned multi-level integrated sensing highway infrastructure monitoring and early warning system. The specific process has been described above and will not be repeated here.

[0145] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0146] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-level integrated highway infrastructure monitoring and early warning system, characterized in that: The end-side optimization deployment module, the fiber-video dual-channel transmission module, the fiber physical field inversion module, the fiber-video redundancy perception module, and the multi-source fusion dynamic risk management module are included. The output end of the end-side optimization deployment module is connected with the fiber signal input end and the video signal input end of the fiber-video dual-channel transmission module, for sending the collected fiber sensing signals and video monitoring data to the fiber-video dual-channel transmission module; the fiber signal output end of the fiber-video dual-channel transmission module is connected with the input end of the fiber physical field inversion module, and the video signal output end is connected with the video data input end of the fiber-video redundancy perception module, for transmitting the processed fiber sensing signals and video monitoring data respectively; the output end of the fiber physical field inversion module is connected with the physical parameter input end of the fiber-video redundancy perception module, for converting the fiber sensing signals into physical parameter information and sending the physical parameter information to the fiber-video redundancy perception module; wherein the physical parameter information includes stress, strain, vibration and temperature parameter information; the output end of the fiber-video redundancy perception module is connected with the input end of the multi-source fusion dynamic risk management module; the fiber-video redundancy perception module includes a deep disease identification unit, an apparent disease identification unit, a disease matching unit and a traffic environment monitoring unit, wherein: The deep disease identification unit detects deep diseases based on the physical parameter information output by the fiber physical field inversion module through a KAST-Net neural network model; The apparent disease identification unit detects apparent diseases based on the video monitoring data output by the fiber-video dual-channel transmission module through a YOLOv12 target detection model; The disease matching unit is configured to complete disease matching and update abnormal probability, and send the matched disease data to the multi-source fusion dynamic risk management module; the disease matching unit adopts a correlation matching probability model considering position error, maps the pixel coordinates of the detected apparent diseases to the camera coordinate system through distortion correction and camera intrinsic parameter mapping, projects them into the structure reference coordinate system combined with the camera extrinsic parameter, and maps the point coordinates of the detected deep diseases along the fiber length to the three-dimensional space coordinates in the structure reference coordinate system combined with the actual route of the fiber layout in the construction drawing through the piecewise linear interpolation method; the spatial coordinate matching of the deep diseases and the apparent diseases is completed through the Euclidean distance calculation and the Hungarian algorithm, and the abnormal probability is updated; The traffic environment monitoring unit identifies the weather state through a ConvNeXt model based on the video monitoring data, and counts the traffic flow through a TransTrack model; The fiber-video redundancy perception module sends the matched disease data, weather state and traffic flow data to the multi-source fusion dynamic risk management module; the multi-source fusion dynamic risk management module realizes risk assessment, grade division and early warning release through the received data.

2. The multi-tiered integrated highway infrastructure monitoring and warning system of claim 1, wherein: The end-side optimization deployment module comprises a fiber sensing layout unit and a video monitoring layout unit; the fiber sensing layout unit is laid along the existing infrastructure of the highway to form a continuous linear sensing network, and the output end of the fiber sensing layout unit is connected with the optical fiber signal input end; the video monitoring layout unit adopts a hierarchical, fixed-point and cross-coverage deployment strategy to set high-resolution cameras according to the different structural characteristics and risk distribution of slopes, bridges, tunnels and main line sections, and construct a discrete point-like sensing network; the output end of the video monitoring layout unit is connected with the video signal input end.

3. The multi-tiered integrated highway infrastructure monitoring and warning system of claim 2, wherein: The fiber-video dual-channel transmission module comprises a fiber signal transmission unit and a video signal transmission unit; the input end of the fiber signal transmission unit is connected with the fiber sensing layout unit; the fiber signal transmission unit receives the fiber sensing signal and performs modulation, combining amplification and relay amplification processing; the output end of the fiber signal transmission unit is connected with the input end of the fiber physical field inversion module, and the processed fiber sensing signal is sent; the input end of the video signal transmission unit is connected with the video monitoring layout unit, and the output end of the video signal transmission unit is connected with the video data input end of the fiber-video redundant sensing module, for sending the processed video monitoring data.

4. The multi-tiered integrated highway infrastructure monitoring and warning system of claim 3, wherein: The fiber physical field inversion module comprises a signal splitting unit and a physical parameter conversion unit; the input end of the signal splitting unit is connected with the output end of the fiber signal transmission unit, receives the fiber sensing signal and splits the signal into a main optical path signal and a sensing optical path signal; the sensing optical path signal output end of the signal splitting unit is connected with the input end of the physical parameter conversion unit; the output end of the physical parameter conversion unit is connected with the physical parameter input end of the fiber-video redundant sensing module; the physical parameter conversion unit can be used for photoelectric conversion, noise filtering, amplification and parameter calculation on the sensing optical path signal.

5. The multi-tiered integrated highway infrastructure monitoring and warning system of claim 4, wherein: The input end of the deep disease identification unit is connected with the output end of the fiber physical field inversion module, the output end of the deep disease identification unit is connected with the first input end of the disease matching unit, for receiving the physical parameter information and detecting the deep disease through a neural network model, and sending the deep disease information to the disease matching unit; The input end of the apparent disease identification unit is connected with the output end of the video signal transmission unit, the output end of the apparent disease identification unit is connected with the second input end of the disease matching unit, for receiving the video monitoring data and detecting the apparent disease through a target detection model, and sending the apparent disease information to the disease matching unit; the output end of the disease matching unit is connected with the disease information input end of the multi-source fusion dynamic risk management and control module; the input end of the traffic environment monitoring unit is connected with the output end of the video signal transmission unit, and the output end of the traffic environment monitoring unit is connected with the environment traffic information input end of the multi-source fusion dynamic risk management and control module, for receiving the video monitoring data and identifying the weather state and counting the traffic flow, and sending the weather state and traffic flow data to the multi-source fusion dynamic risk management and control module.

6. The multi-tiered integrated highway infrastructure monitoring and warning system of claim 5, wherein: The multi-source fusion dynamic risk management and control module comprises a risk scoring unit, a risk grading unit and an early warning issuing unit; the input end of the risk scoring unit is connected with the output end of the fiber-optic-video redundancy perception module, the output end of the risk scoring unit is connected with the input end of the risk grading unit, is used for receiving disease data, weather conditions and traffic flow data, calculating a risk score in combination with location-sensitive information, and sending the risk score to the risk grading unit; The output end of the risk grading unit is connected with the input end of the early warning issuing unit, divides risk levels according to a preset threshold value built-in, and sends risk level information to the early warning issuing unit; The early warning issuing unit generates a differentiated treatment strategy in combination with regional types and sends early warning information to a management end, a field end and a driver-passenger end.

7. A multi-level integrated highway infrastructure monitoring and early warning method based on the multi-level integrated highway infrastructure monitoring and early warning system of any one of claims 1-6, characterized in that, Comprise: S1: deploy fiber-optic sensing devices and video monitoring devices through an end-side optimization deployment module, collect fiber-optic sensing signals and video monitoring data of key areas of highway infrastructure, and send them to a fiber-optic-video dual-channel transmission module; S2: the fiber-optic-video dual-channel transmission module performs low-loss transmission on the received fiber-optic sensing signals and video monitoring data, sends processed fiber-optic sensing signals to a fiber-optic physical field inversion module, and sends processed video monitoring data to a fiber-optic-video redundancy perception module; S3: the fiber-optic physical field inversion module converts the received fiber-optic sensing signals into physical parameter information and sends them to the fiber-optic-video redundancy perception module, the fiber-optic-video redundancy perception module processes the physical parameter information through a KAST-Net neural network model to detect deep diseases, processes the video monitoring data through a YOLOv12 target detection model to detect apparent diseases, completes disease matching and updates abnormal probability through Euclidean distance calculation and the Hungarian algorithm, and simultaneously sends results to a multi-source fusion dynamic risk management and control module through a ConvNeXt model to identify weather conditions and a TransTrack model to count traffic flow; S4: the multi-source fusion dynamic risk management and control module calculates a risk score, divides a risk level and issues a differentiated early warning strategy according to received disease identification results, weather conditions and traffic flow data.

8. The multi-level integrated highway infrastructure monitoring and early warning method according to claim 7, characterized in that, In step S1, the end-side optimization deployment module relies on existing infrastructure along the highway to lay optical cables, forming a continuous linear perception network; for different structural characteristics and risk distribution of slopes, bridges, tunnels and main line sections, a high-resolution camera is set up using a hierarchical, fixed-point and cross-coverage deployment strategy to construct a discrete point perception network.

9. The multi-level integrated highway infrastructure monitoring and early warning method according to claim 7, characterized in that, In step S3, the signal splitting unit of the optical fiber physical field inversion module splits the received optical fiber sensing signal into a sensing optical path signal, the physical parameter conversion unit performs photoelectric conversion, noise filtering, amplification and parameter calculation on the sensing optical path signal, generates physical parameter information and sends it to the optical fiber-video redundancy perception module; the optical fiber-video redundancy perception module detects deep diseases based on the physical parameter information, detects apparent diseases based on the video monitoring data, converts the coordinates of the two types of diseases to a unified coordinate system and completes matching, updates the abnormal probability, and identifies the weather state and counts the traffic volume based on the video monitoring data, sends the disease matching result, weather state and traffic volume data to the multi-source fusion dynamic risk management and control module.

10. The multi-level integrated highway infrastructure monitoring and early warning method according to claim 9, characterized in that, In step S4, the multi-source fusion dynamic risk management and control module combines the disease severity, weather state, traffic volume, location sensitive information and disease abnormal probability to calculate the risk score of each region of the highway; According to the preset threshold, the risk score is divided into different risk levels; Combined with the risk level and the region type, a differentiated treatment strategy is generated and warning information is sent to the management end, the field end and the driver and passenger end.

Images